Publication Record
Evidence current through: Thursday, November 19, 2026, at 1:20 a.m. UTC (8:20 p.m. ET and 5:20 p.m. PT on Wednesday, November 18).
This note summarizes the eight companion notes of the Greyhound Research Microsoft Ignite 2026 series and reads each product at the status Microsoft gave it by the time above. The sources were checked again for this edition, after the companion notes completed their own pre-launch audit. Most agent features announced on November 17, 2026, are in preview or Microsoft’s Frontier program. Developments after that time are not reflected. Times in the text are given in UTC.
Research Note In Two MinutesNote synopsis · AI drafted · Analyst reviewed
Drafted by AI from the full research note and reviewed by a Greyhound Research analyst before publication. The analysis and conclusions are Greyhound Research’s own.
At Microsoft Ignite on November 17, 2026, Microsoft treated AI agents as part of the workforce rather than a feature of its software. Agents now get a name in Microsoft’s directory, a sealed-off machine of their own, a budget drawn from a shared meter, and a choice of Microsoft’s own models. Copilot became the front door to Microsoft 365, security agents began closing alerts on their own, and a released tool now helps companies leave Salesforce. Much of what decides cost and risk is still in preview or Microsoft’s early-access Frontier program.
Greyhound Research’s position is that Ignite 2026 changed who holds the control. In every area its eight-note series examines, Microsoft shipped a capable tool and left the rule that tool enforces for the buyer to write. When agents join the workforce, written authority becomes the control: who may enroll, spend, act, and leave, set down by the buyer before the agent starts work. Greyhound Research reads this as a normal stage of a platform shift. The capability has arrived ahead of the terms, as it did with software as a service and public cloud, and the buyers who fare best tend to adopt what is released and write their own rules for the rest.
Three findings carry the argument. First, Microsoft has written the first steps of an agent’s authority, its name and its machine, and not yet the rule at each tool call or the power to stop a rival’s agent. Second, the meter now sits beside every seat price, and several of its rates are unpublished. Third, Microsoft now meters the spend, scores the value, grades its own security agents, and sells its own models, so buyers need records Microsoft does not mark.
The note sets out a ten-move Greyhound CXO Runbook for the whole series, with six exhibits, five tables, and six questions for Microsoft. None of the moves waits on Microsoft. The first three are simple to state. Sign on general availability terms and enroll named groups only. Write the spending policy before the meter runs. Register every agent as a privileged account and map where it can be stopped.
The evidence has limits. This note summarizes eight companion notes and does not repeat their evidence, which each one sets out in full. Most agent features are previews, and Microsoft’s pages word some statuses differently. Greyhound Research names the signals that would change its view within twelve months. Until they arrive, the advice holds: use what Microsoft has released, and write down the rules for everything an agent may do.
In This Research NoteSection guide · AI drafted · Analyst reviewed
Sixteen sections in seven parts, each with a short summary. Select a title to go straight to that section. Each section opens with a Greyhound Spotlight, the one fact in it a reader would not expect, and eleven end with a block named The Deeper Read, which holds the fuller argument and the passages it rests on. Select any line in that block to open its full text.
- Part 1What Ignite Changed
- 1.1What Ignite Changed: Agents Moved From Feature To Workforce
Microsoft gave agents names, machines, budgets, and models of their own. Most of what decides cost and risk is still in preview or the Frontier program.
- 1.2What Microsoft Announced: Ten Announcements, Ranked By What They Change
Greyhound Research’s top ten by consequence for buyers, with each one’s status. Copilot as the front door ranks first. Most of what decides cost is in preview.
- 1.3A Year On: From Ignite 2025’s Previews To Ignite 2026’s Price List
At Ignite 2025 Agent 365 was in the Frontier program and agents were being named. A year later they have prices, budgets, and a count of 50 million.
- Part 2The Through-Line
- 2.1The Through-Line: The Control Moves To What The Buyer Writes
In each area the series examines, Microsoft shipped a tool and left the rule it enforces to the buyer. Eight notes name eight controls, and each is a written document.
- 2.2The Series: Eight Notes, Eight Controls
The eight companion notes, each with its control and its readers, and the order in which to read them by role.
- Part 3The Eight Notes In Brief
- 3.1The Front Door And The Bill: Copilot And Agent Economics
Copilot’s new front door reaches every user this winter while its agents stay in Frontier. Copilot Credits are one currency but not one bill, and the spending policy decides the cost.
- 3.2The Name, The Machine, And The Seal: Identity, Infrastructure, And Confidential Computing
Three notes follow an agent from its directory name to its own machine and its sealed data. Microsoft has written the name and the machine, and not yet the limit or the stop.
- 3.3The Defender, The Model, And The Exit: Perception, MAI, And Activate
Security agents now close alerts on their own, Microsoft sells its own models beside its partners’, and a released tool prices the way out of Salesforce. Each needs a record Microsoft does not keep.
- Part 4The Risks And The Counter-Argument
- 4.1Risks To Watch: Where The Series’ Controls Can Fall Short
Five risks that run across the eight notes, from previews drifting into production to the seller marking its own work, each with an early warning and an owner.
- 4.2The Strongest Objection: Microsoft Will Write The Controls Itself
The fair counter-case: Microsoft ships more control tooling than its rivals, and one suite simplifies buying it. Both hold. The claim that the tools are the control does not.
- Part 5What To Do, By Role
- 5.1The Greyhound CXO Runbook: Ten Moves Across The Series
Ten series-level moves for the CIO, CFO, CISO, data protection officer, procurement, general counsel, change and adoption leads, and the chief risk officer, each naming the assumption it corrects and who decides.
- 5.2Questions For Microsoft: What Buyers Will Ask Next
Six series-level questions only Microsoft can answer, on release terms, rates, enforcement, data, exit, and audit. Each is tied to a Runbook move.
- Part 6The Final Greyhound Standpoint
- 6.1Microsoft Ignite 2026: The Final Greyhound Standpoint
Agents joined the workforce, and written authority is the control. The limits of the evidence, and the signals that would change the view within twelve months.
- Part 7Appendix, Sources, And Disclosures
- 7.1Appendix: The Evidence Register, Source By Source
Every source behind this research note, with when it was read and where it is used. Passages quoted in The Deeper Read are not repeated here.
- 7.2Sources And References: Every Document Behind Every Claim
Twenty-three sources in five classes: Microsoft’s announcements, a partner announcement, media, the eight companion notes, and earlier Greyhound Research notes.
- 7.3Disclosures: Independence, Financial, Intended Use, Methodology, Authoring, And Copyright
Independence first, then the Financial Disclosure, intended use, research method and its limits, authoring, and copyright terms.
Where To Start, By RoleReader guide · AI drafted · Analyst reviewed
Find your role and read its sections in the order shown. Each line takes you straight to the section.
Chief information officer
Chief financial officer
Chief information security officer
Data protection officer
Head of procurement
Chief risk officer
Change and adoption lead
General counsel
Part 1What Ignite Changed
What Ignite Changed: Agents Moved From Feature To Workforce
Greyhound SpotlightMicrosoft’s chief executive said the week was about two things. The second was building the capability to deal with the consequences of fast-growing AI, and every note in this series is about that second thing.
At Ignite in San Francisco on November 17, 2026, Satya Nadella, Microsoft’s chief executive, called Copilot “the new OS for work.” The phrase carried the week. Copilot, Microsoft’s AI assistant, became the starting point for work in Microsoft 365, with Word, Excel, and PowerPoint inside it. Around it, Microsoft treated agents, AI programs that carry out tasks on a person’s behalf, less like features and more like staff.
Agents now get most of what a new employee gets. They get a name in Entra, Microsoft’s directory of people and systems. They get a sealed-off machine for each task, a budget drawn from Copilot Credits, Microsoft’s usage-based currency, and a choice of Microsoft’s own AI models beside those of OpenAI and Anthropic. Security agents in Microsoft Perception now close alerts without asking anyone first. Microsoft’s own summary put the stakes this way: as agents move from answering questions to taking action, trust becomes the deciding factor.
The workforce comparison is about governance, not about what an agent is. Greyhound Research said in January 2026 that service agents should be monitored “like they’re privileged employees,” and the comparison here borrows the same discipline: a name, a sponsor, narrow rights, and a way to remove them. It does not make an agent an employee, and agents should not be bought, priced, or credited as if they were.
The keynote also opened with an apology. Days after a Microsoft 365 outage, Nadella said that Microsoft takes the trust customers place in its services extremely seriously and has work to do on resilience. By Greyhound Research’s count, that outage was the fourth multi-service Microsoft 365 incident of 2026. Greyhound Research said on November 13, and repeated in its November 15 note on that outage, that enterprises have priced Microsoft 365 as a license rather than as a single point of failure. Agents raise the same question one level up.
What this changes is the unit a buyer governs. For three decades, enterprise software has been governed product by product, through a license count and the settings that switch features on. Ignite moves part of that work down to the individual agent, each with its own name, machine, and budget, so the familiar joiner, mover, and leaver routine of human resources now needs a software counterpart. Agent numbers are likely to grow faster than any review cycle built for staff. The effect most coverage misses is the leaver. When the employee who sponsored an agent changes role or leaves, the agent stays, and its rights stay with it unless someone has written down who removes them.
Exhibit 1 groups the week’s news by what a buyer can sign now, what is still on test terms, and what is not yet written. Generally available means fully released, with support and service terms. Frontier is Microsoft’s early-access route for previews.
Greyhound Research · Exhibit 1 · What Ignite 2026 released, and what it left open
The products shipped. The rules for agents did not
Microsoft’s Ignite 2026 announcements grouped by what a buyer can sign today, what is still on test terms, and what is not yet written, November 19, 2026.
Greyhound Research’s judgment: commit only on the first card, pilot the second, and treat the third as the buyer’s to write.
Exhibit 1 groups Microsoft’s Ignite 2026 announcements into three cards. Generally available: Home and Office in Copilot, Microsoft Perception, Microsoft-Decision-1.1 with MAI Code 2 live in GitHub Copilot, and the Salesforce migration in Dynamics 365 Activate. In preview or the Frontier program: Autopilot, Workspaces, and Today, Agent 365 Cost Management, connected platforms for rivals’ agents, and Agent Grid, the AKS sandbox extension, and the Confidential Clean Room in limited preview. Not yet written: several Copilot Credits rates and a Perception rate card, prices for Licensed Weights and Activate, HIPAA coverage and three ISO standards for Frontier features, whether agent memory sits inside the seal or controls are enforced on rivals’ platforms, and the confidential computing items Microsoft describes as still in development. The grouping is Greyhound Research analysis.
The first card is real and usable today. The second holds most of the agent features that justify the week’s claims, and they sit on preview terms. The third card matters most, because it lists what no one has written yet: rates, prices, coverage, and the boundaries of what an agent can reach. Those gaps do not wait for Microsoft to close them. Each one is a place where the buyer’s own written rule decides the risk.
Greyhound Standpoint:
At Greyhound Research, we believe Ignite 2026 marks the point at which Microsoft began to sell agents as a workforce rather than as a feature, and that the first question it raises is about mandate, not technology. The reason lies in what Microsoft attached to each agent. A directory name, a machine, a budget, and a choice of model resemble what a company gives a new member of staff, and each one opens a cost line, a risk line, or both. A feature is governed by the setting that switches it on. A worker is governed by a written mandate that says what it may touch, spend, and decide, and who answers when it gets something wrong.
Practitioners have seen this before. When personal computers, email, and then cloud services reached employees, use spread well ahead of the policies that later governed it, and catching up cost more with each month of delay. Microsoft deserves credit for putting trust at the center of its message and for opening the keynote by addressing the outage. Much of the agent estate is still in preview, so the mandate may change as terms are published. CIOs should table a one-page agent mandate at the next architecture review, naming who may deploy an agent, what it may reach, what it may spend, and who can stop it, before any agent moves past pilot.
The Deeper Read · About 3 minutes
Select a question to open the full answer
1What did Microsoft say the week was about?
Two things. The keynote, posted by Microsoft as an edited transcript, says the week was about capitalizing on capabilities growing exponentially, and building the capability to deal with their consequences. The same text calls Copilot “the new OS for work.” On stage, Nadella added that apps are “just plugins to Copilot.”
Trust as the deciding factor. Microsoft’s Source summary of the event names four themes: human ambition, intelligence, trust, and infrastructure. Its trust section says that trust becomes the deciding factor as agents move from answering questions to taking action.
What the two framings tell a buyer. The keynote and the Source summary put the same week in two ways. One speaks of capability and its consequences, the other of trust as the deciding factor. Read together, they show Microsoft asking to be judged on trust, which is welcome, and they give buyers a simple test for each announcement: does it add capability, or does it help deal with the consequences? On Greyhound Research’s reading, most of the released items add capability, and most of the consequence work, the rules, rates, and terms, is still to come.
In the source’s own words
- The Official Microsoft Blog, Nadella keynoteThis week is about two things: First is how we capitalize on new capabilities that are growing exponentially.Source
- The Official Microsoft Blog, Nadella keynoteAnd second is how do we build the capability to deal with the consequences of these exponentials.Source
- The Official Microsoft Blog, Nadella keynoteCopilot is the new OS for work.Source
- Microsoft Source, November 17, 2026As agents move from answering questions to taking action, trust becomes the deciding factor.Source
2How did the outage shape the keynote?
An apology first. Nadella acknowledged the previous week’s outage. He said Microsoft knows the trust customers place in its services and takes that responsibility extremely seriously, and that it has work to do on resilience. Greyhound Research’s outage note, which counts the incident as the fourth multi-service Microsoft 365 incident of 2026, found that consolidating on one suite remains right for most enterprises while the standard terms around it no longer suffice.
Consolidate, and keep the record independent. Greyhound Research has also warned, in its July 2025 note on Copilot’s Researcher and Analyst agents, that as agents become embedded in business-critical workflows the enterprise becomes progressively dependent on them, and “renegotiation leverage is materially diminished.” The two views fit together, and this series carries both. Consolidating the suite remains sensible for most enterprises, on the condition that the evidence of what agents did, the approval of what they may do, and the means to recover sit outside the suite being governed. The same condition explains why Copilot as the new OS for work is treated here as a dependency to govern as well as a convenience.
How the outage advice has moved. In its June 2025 outage note, Greyhound Research wrote that no single vendor can serve as the sole source of continuity, and advised failover messaging tools that integrate with Microsoft’s identity service. The November 2026 note keeps the first point and narrows the second. It advises against a second suite held in reserve, and asks instead for a rehearsed fallback for the few processes that cannot stop, one that shares nothing with the suite, including its sign-in. A fallback that signs in through the suite shares its outage, which is why the advice changed.
In the source’s own words
- Satya Nadella, Microsoft’s chief executive, Ignite keynoteWe know the trust you place in our services and take that responsibility extremely seriously.Source
- Satya Nadella, Microsoft’s chief executive, Ignite keynoteI know we have work to do to provide that resilience that you can count onSource
- Greyhound Research, November 13, 2026, as quoted in its Microsoft 365 outage note of November 15Enterprises have priced Microsoft 365 as a licence and never as a single point of failure.Source
- Greyhound Research, Microsoft 365 outage noteConsolidation remains the right strategy for most enterprises. The standard terms around it are no longer enough.Source
- Greyhound Research, Microsoft 365 outage noteGreyhound Research does not recommend a second suite held in reserve.Source
- Greyhound Research, June 18, 2025 outage noteNo single vendor, regardless of reputation or investment, can serve as a sole source of continuity when their orchestration layer is a singular point of failure.Source
Part 1What Ignite Changed
What Microsoft Announced: Ten Announcements, Ranked By What They Change
Greyhound SpotlightFour of the ten announcements that matter most to buyers are generally available at their core. The other six, including every item that sets cost or identity, are mostly in preview or Frontier.
Microsoft’s Book of News, its official list of Ignite announcements, covers every product line the company sells. Greyhound Research ranked them by one test: how much each changes a buyer’s budget, contracts, or risk in the next twelve months. Status matters as much as substance. A preview is a test release without full service terms, so a preview cannot carry a production commitment however large the keynote made it. By Greyhound Research’s count of the 59 announcement headings in the Book of News, 10 cover only generally available items, 19 cover only previews or Frontier items, and 28 mix the two. The other two give no status for some or all of their items.
Greyhound Research reads the heading count as more telling than any single launch. With 28 headings mixing released and preview items, one product name can now cover both, and a purchase approved by name can carry a preview inside it. In earlier cycles a product was usually either shipping or not, and the release note settled the question. Here status sits at the level of the feature, and it is likely to keep moving month by month as the December dates arrive. The second-order effect is administrative. Internal approvals, risk registers, and contract schedules that list products by name will need to list features, or they will record a status the vendor has already changed.
Table 1 sets out the ten, in order, with the status Microsoft gave each and the companion note that covers it. Nine of the ten are covered by a companion note in this series.
Greyhound Research · Table 1 · The top ten, by buyer consequence
Ten announcements, and most of what decides cost is in preview
Greyhound Research’s ranking of Microsoft’s Ignite 2026 announcements by consequence for enterprise buyers, with status as Microsoft gave it, November 19, 2026.
Judge each row by its status, not its rank. Only the four released rows are ready for a contract.
| Rank | Announcement | Status at Ignite | What changes for the buyer | Companion note |
|---|---|---|---|---|
| 1 | Copilot as the new OS for work | Home and Office in Copilot generally available. Autopilot, Workspaces, and Today in Frontier | Where every user starts work moves this winter | Copilot |
| 2 | Agent 365 Cost Management and Business Analytics | Cost Management in preview. Business Analytics in Frontier | Spend and value reporting become Microsoft’s own tools | Economics |
| 3 | Copilot Credits across more services | Live for some services. Several rates not published | One meter now sits beside every seat price | Economics |
| 4 | Connected platforms for rivals’ agents | Rolling out, completing in December | Microsoft bids to be the directory of every agent | Identity |
| 5 | Microsoft Perception and the E5 and E7 SIEM offer | Generally available. Vulnerability shielding in preview | Security agents now close alerts on their own | Perception |
| 6 | Dynamics 365 Activate and Copilotforce | Salesforce migration generally available. SAP ECC in private preview. Copilotforce in preview | Renewal leverage with Salesforce and SAP changes | Activate |
| 7 | MAI models and MAI Licensed Weights | Code 2 live in GitHub Copilot. Decision-1.1 generally available. Licensed Weights coming | Microsoft becomes supplier and rival to its model partners | MAI |
| 8 | Local and shadow agent governance | Local agent discovery generally available. Governance in preview | Agents on employees’ computers become managed assets | Identity |
| 9 | Azure agent infrastructure and Cobalt 200 | Cobalt 200 generally available. Agent Grid and AKS sandboxes in preview | Agents get machines, network rules, and memory of their own | Infrastructure |
| 10 | Microsoft IQ and Azure HorizonDB | Both generally available. No new SKU for Microsoft IQ | The context layer ties agents to Microsoft’s data estate | Not covered |
Table 1 ranks ten Ignite 2026 announcements by consequence for buyers, with each one’s status and the companion note that covers it. First, Copilot as the new OS for work, partly generally available and partly in Frontier. Second, Agent 365 Cost Management and Business Analytics, in preview and Frontier. Third, Copilot Credits across more services, with several rates unpublished. Fourth, connected platforms for rivals’ agents, completing in December. Fifth, Microsoft Perception, generally available. Sixth, Dynamics 365 Activate and Copilotforce. Seventh, MAI models and Licensed Weights. Eighth, local and shadow agent governance. Ninth, Azure agent infrastructure and Cobalt 200. Tenth, Microsoft IQ and Azure HorizonDB, not covered by a companion note. The ranking is Greyhound Research analysis.
Copilot as the front door ranks first because it reaches every licensed user this winter, whatever a company decides about agents. The economics items rank next because the meter now touches every product. The four rows released at their core are the ones a buyer can sign today: Perception, the MAI coding models live in GitHub Copilot with the generally available Decision-1.1, Activate’s Salesforce migration, and Microsoft IQ. The rest belong in a plan, not a contract.
Greyhound Standpoint:
Greyhound Research’s reading is that Ignite 2026 released the parts of the agent story a buyer can see and kept in preview most of the parts a buyer will rely on. That gap should set the order of work. Four of the ten items that matter most are released at their core: Perception, the MAI models live in GitHub Copilot with the Decision-1.1 routing model, the Salesforce migration in Activate, and Microsoft IQ. The spending controls, the reach into rivals’ agents, and the machinery for running agents remain on test terms. A preview carries no full service terms, so a commitment built on one rests on terms that do not yet exist.
The pattern is a familiar one. Platform vendors have long announced the whole vision at a flagship event and released it in stages, and buyers who signed on the vision ahead of the release could end up paying for what they could not yet use. Microsoft labels its previews clearly, which is to its credit, and several open items carry December dates. If those dates hold with full terms attached, the gap will narrow quickly. CIOs and procurement leads should hold each item in Table 1 to its status at the next pilot gate, adopting the released rows on their merits, budgeting the previews as plans, and accepting no production commitment that rests on a keynote status.
The Deeper Read · About 2 minutes
Select a question to open the full answer
1How was the ranking made?
By buyer consequence. Greyhound Research read Microsoft’s Book of News, the keynote, and the product blogs, and ranked each announcement by its effect on a buyer’s budget, contracts, or risk within twelve months. The statuses are as Microsoft gave them. Where Microsoft’s pages disagree, the companion notes record both readings.
What the statuses say. The Book of News lists Perception, Microsoft IQ, and Azure HorizonDB as generally available, and calls the Salesforce migration in Dynamics 365 Activate generally available. Microsoft IQ comes, in Microsoft’s words, with no new SKU or subscription. Cobalt 200 virtual machines, Microsoft’s own Arm-based servers, are generally available too.
In the source’s own words
- Microsoft’s Book of News, Microsoft PerceptionMicrosoft Perception is now generally available,Source
- Microsoft’s Book of News, Dynamics 365the Salesforce migration experience is now generally available.Source
- Microsoft’s Book of News, Microsoft IQMicrosoft IQ, now generally available,Source
- Microsoft Source, November 17, 2026Cobalt 200 virtual machines, generally available today, deliver up to 50% better performance than the previous generation.Source
2What did not make the ten?
Confidential computing. The Book of News lists three confidential computing offerings as in preview, under a heading on protecting highly sensitive data. Microsoft’s more detailed confidential computing post, by Vikas Bhatia, reads more narrowly: one limited preview and two items still in development. It describes the multi-GPU item as a technology showcase whose confidential VMs and containers remain in development, and the enclaves as planned for a future release. Only the clean room, first previewed in June, is in limited preview. This note follows the detailed post, as the companion notes do. None of the three has a price or release date, so the news ranks below the ten for most buyers. For banks, insurers, and health groups it may matter more than anything on the list, and it has its own companion note.
What the gap between the two pages tells a buyer. The Book of News is a summary written for the whole event, and the product team’s post is written for the people who will build on the product. When the two differ, the more detailed page is usually the safer guide, and the difference is itself a sign of how early the product is. A buyer’s status check should cite the most specific Microsoft page available, with the date it was read, rather than the event summary.
Silicon and Windows. Maia 300, Microsoft’s next AI chip, was highlighted with a claim of 40 percent more tokens per watt than leading GPU systems, a company figure this note does not use. Windows security changes, including a Windows Endpoint Security Platform, now generally available, that gives approved security products a safer way to protect devices, matter to endpoint teams and sit outside this series.
In the source’s own words
- Microsoft’s Book of News, AzureAzure expands confidential computing for protecting highly sensitive dataSource
- Vikas Bhatia, Microsoft, Azure confidential computing at Ignite 2026VR72 confidential VMs, AKS and Kata-based confidential containers, and Microsoft Foundry integration remain in development.Source
- Vikas Bhatia, Microsoft, Azure confidential computing at Ignite 2026Azure Confidential Clean Rooms, in limited preview, applies this thinking across organizations.Source
Part 1What Ignite Changed
A Year On: From Ignite 2025’s Previews To Ignite 2026’s Price List
Greyhound SpotlightAt Ignite 2025, Agent 365 was an early-access preview. By May 2026 it had a price of USD 15 per user, and Microsoft now says more than 50 million agents have been registered in it.
A year ago, Microsoft’s Ignite was about naming agents. Agent 365, its tool for keeping track of a company’s agents, arrived through the Frontier program, and agents built in Copilot Studio began to get an Entra Agent ID. Microsoft’s post that week spoke of organizations becoming Frontier Firms. Two of its new context services, Fabric IQ and Foundry IQ, were in preview, and Azure HorizonDB, a new database, was in private preview.
Twelve months on, the same products have prices and general availability. Microsoft announced in March that Agent 365 and Microsoft 365 E7, its top suite, would be generally available on May 1, at USD 15 and USD 99 per user. At Ignite 2026, Microsoft IQ, the umbrella over its context services, and Azure HorizonDB became generally available. Microsoft says organizations have registered more than 50 million agents in Agent 365 since its May launch.
The pace is the news here. Microsoft took Agent 365 from an early-access preview to a priced, generally available product in less than a year, between Ignite 2025 and May 1, and buyers should plan for the next steps to move at a similar speed. Most enterprise governance runs on an annual cycle of budgets, audits, and renewals, so the vendor’s calendar now runs quicker than the buyer’s. The consequence most coverage misses is the estate already built. Agents created during the preview year, under test terms, now sit inside paid products, and the first governance task is to find and classify them before planning new ones.
Exhibit 2 places the steps between the two events on a month grid, from the first prices to the security agents.
Greyhound Research · Exhibit 2 · The year between two Ignites
In twelve months, agents went from preview to price list
Microsoft’s steps from Ignite 2025 to Ignite 2026, by month, from the first price for agent management to the general availability of its security agents.
Microsoft delivered on its year. Buyer governance now has to keep the same pace.
Exhibit 2 places Microsoft’s steps between Ignite 2025 and Ignite 2026 on a month grid from December 2025 to November 2026, and shows six months with a step from preview to product. On March 9, Microsoft set prices for Microsoft 365 E7 and Agent 365. On May 1, both became generally available, at USD 99 and USD 15 per user. On June 16, Cowork became generally available. On July 27, Project Perception was introduced. In September, Microsoft announced Home in Copilot. On November 17, Ignite opened days after a Microsoft 365 outage. A note beneath records the starting point: at Ignite 2025, Agent 365 was in the Frontier program and Fabric IQ and Foundry IQ were in preview.
The pattern is a move from naming to billing. In 2025 the question was whether an agent could be found and listed. In 2026 it is what an agent costs, what it may do, and who answers for it. Microsoft’s own summary now describes last year’s theme as Frontier Transformation, and this year’s news as its next milestone. The buyer’s work has moved with it, from inventory to policy.
Greyhound Standpoint:
Greyhound Research’s view is that Microsoft used the year well, and that the move from preview to price list changes the buyer’s job from watching to governing. Products that were early-access previews at Ignite 2025 now carry prices and general availability, and Microsoft reports more than 50 million agent registrations in Agent 365 since May. Greyhound Research called Entra Agent ID a policy control plane for agent computing in its Build 2025 note, and Microsoft has since extended it to rivals’ agents. That is steady delivery.
The mechanism behind the shift is the meter. A preview costs little to watch. A released product with a per-user price and a usage meter creates a commitment that grows each month whether or not anyone has decided what it is for. The same arc ran through the early years of software as a service, when trial seats became enterprise contracts and ownership stayed with IT long after the spending had spread across departments. Fifty million registrations measures reach rather than governed use, and a published definition would make it more useful. CFOs, CISOs, and general counsel should each take written ownership of a part of the agent estate at the next budget cycle, so that the meter, the access, and the terms do not rest with the IT team alone.
The Deeper Read · About 2 minutes
Select a question to open the full answer
1What did Ignite 2025 announce?
Agents named, not yet billed. The Ignite 2025 Book of News says Agent 365 was available in the Microsoft 365 admin center with the Frontier program, and that agents built in Copilot Studio would get a Microsoft Entra Agent ID. It lists Fabric IQ and Foundry IQ in preview and Azure HorizonDB in private preview. Frank X. Shaw’s post that week described Agent 365 as a way to observe, manage, and secure agents.
In the source’s own words
- Microsoft’s Book of News, Ignite 2025Agent 365 is available now in Microsoft 365 admin center with the Frontier program,Source
- Microsoft’s Book of News, Ignite 2025Agents built in Copilot Studio will now get a Microsoft Entra Agent ID.Source
- Microsoft’s Book of News, Ignite 2025Foundry IQ, also in preview, will be the next generation of retrieval-augmented generation (RAG)Source
- The Official Microsoft Blog, November 18, 2025Microsoft Agent 365 enables you to observe, manage and secure your AI agents,Source
2What changed by Ignite 2026?
Prices, then volume. Judson Althoff, CEO of Microsoft’s commercial business, set general availability for Agent 365 and Microsoft 365 E7 on May 1, 2026, and E7 bundles Microsoft 365 E5, Microsoft 365 Copilot, and Agent 365. Microsoft’s Agent 365 blog gives the 50 million as registrations since the May launch. The companion note on agent identity finds three wordings of the figure and no published definition, so it counts registrations, not governed agents.
How the view has moved. Greyhound Research said in March 2026 that “there is no mature control plane that governs how agents act,” what they access, or how their actions can be reversed or contained. Since then, Agent 365 and E7 have reached general availability, and Microsoft has extended Entra Agent ID to rivals’ agents and to agents on employees’ computers. The view has moved in part. For Microsoft’s own agents and for managed devices, a control plane now exists. For agents on rivals’ platforms, the rule at each tool call and the power to stop an agent are not yet written, which is where this series places the remaining gap.
What three wordings of one figure tell a buyer. A registration count measures how many agents have been listed, not how many are in use or governed. That the 50 million is worded three ways, with no published definition, does not make it wrong. It means the figure cannot yet serve as a benchmark, either for a buyer’s own estate or against a rival’s. A company that wants a number it can compare over time will need to define its own, for example agents active in the past month with a named sponsor.
In the source’s own words
- The Official Microsoft Blog, March 9, 2026General availability of Agent 365 on May 1 for $15 per userSource
- The Official Microsoft Blog, March 9, 2026General availability of the new Microsoft 365 E7: The Frontier Suite on May 1 for $99 per userSource
- Microsoft Agent 365 Blog, Ignite 2026Organizations have registered more than 50 million agents in Microsoft Agent 365 since its May 2026 launch.Source
- Microsoft Source, November 17, 2026One year ago at Microsoft Ignite, we introduced Frontier Transformation.Source
Part 2The Through-Line
The Through-Line: The Control Moves To What The Buyer Writes
Greyhound SpotlightEvery one of the eight controls this series names is a document the buyer writes: a list, a policy, a map, a rule, a test, an audit, a contract term, or a plan.
Greyhound Research read Ignite 2026 in eight companion notes, one for each area where a buyer’s decisions change. Each note ends with the same kind of finding. Microsoft has built a capable tool. The tool enforces a rule. The rule is not in Microsoft’s product, its price list, or its terms. It has to be written by the customer, and it is now the thing that decides the risk.
The pattern holds from the front door to the exit. Copilot’s agents reach only the users a company enrolls in Frontier, so the enrollment list is the control. Copilot Credits are predictable only under a spending policy. A rival’s agent can be stopped only where its credentials are issued, so the map of those places matters more than the directory. The same is true of authority, proof, audit, model choice, and the plan to leave. Greyhound CIO Pulse 2026, across more than 1,000 enterprise technology leaders, found that 84% have shifted their focus from AI adoption to AI accountability. Accountability needs a written rule to account against.
What is new is the kind of rule being asked for. Earlier enterprise controls mostly governed access: who could open a system, see a record, or approve a payment. Agent controls govern action, what software may do on its own once inside, and most governance frameworks were written before that question arose. The 84% figure suggests the demand for accountability is already ahead of the documents that would support it, and Greyhound Research expects that gap to surface first in audit committees and internal audit plans. The second-order effect is that a written rule is also a record. Once it exists, auditors and regulators can ask for it, and the company can be measured against what it wrote.
Exhibit 3 holds the eight notes against three tests: whether the product is released today, whether its price and terms are written, and whether Microsoft’s tool settles the control on its own.
Greyhound Research · Exhibit 3 · One through-line, eight controls
In every area, the control is something the buyer writes
The eight companion notes of Greyhound Research’s Microsoft Ignite 2026 series, each with the control it names, against three tests, November 19, 2026.
This is not a gap in Microsoft’s tools. It is work only the buyer can do, and it should be staffed now.
Exhibit 3 is a matrix of the eight companion notes against three tests, and shows that in no area does Microsoft’s tool settle the control alone. Released today: yes for Perception, in part for Copilot, economics, identity, infrastructure, MAI models, and Activate, and not yet for confidential computing, where only the clean room is in limited preview. Price and terms written: in part for economics, where seat prices are published and several credit rates are not, and not yet for the other seven. Microsoft’s tool settles it alone: no for all eight, whose controls are Frontier enrollment, the spending policy, enforcement, authority, attestation, the audit, model choice, and the exit plan. The marks are Greyhound Research analysis.
Only one area, Perception, is generally available at its core, and only one, agent economics, has price and terms written in part. In none does the tool settle the control alone. That is not a charge against Microsoft. A spending limit, a directory, or an attestation service can only enforce a rule someone has written. The finding is that the rule-writing has moved to the buyer, and many buyers are only beginning to staff it.
Greyhound Standpoint:
Greyhound Research’s position is that Ignite 2026 has a single through-line. When agents join the workforce, written authority becomes the control, and that authority is a set of documents the buyer writes and owns: who may enroll, spend, act, and leave, set down before the agent starts work. Every Microsoft tool read for the series enforces a setting, whether a spending limit, a Conditional Access policy, or an attestation check, and a setting is only as sound as the rule behind it. Where no rule is written, the default decides, and defaults are designed for the vendor’s general case rather than the buyer’s particular risk.
Experienced buyers will recognize the lesson from the outsourcing era. Contracts moved the work to a provider, and the retained team that wrote and policed the service terms came to matter as much as the contract itself. Microsoft deserves credit for building tools that can enforce such rules and for its stated commitment to transparency. This note relies on the evidence of eight companion notes, and a documented rule at the tool call from Microsoft would move part of the control back into the product. Boards should ask, at their next risk review, to see six documents by name: the Frontier enrollment list, the spending policy, the agent register with its stop points, the audit sample, the model policy, and the exit plan.
The Deeper Read · About 2 minutes
Select a question to open the full answer
1Why written, and why the buyer?
A rule needs an author. Each Microsoft tool read for the series enforces a setting: a spending limit, a Conditional Access policy, an attestation check, or a review queue. The companion note on agent economics finds that a spending limit is a ceiling, not a budget. The note on Azure agent infrastructure finds that the layer deciding what an agent may do was left to be written. In both, the setting is only as good as the rule behind it.
Accountability. Greyhound Research’s November 5 note on agent front doors cited the Greyhound CIO Pulse 2026 finding that 84 percent of more than 1,000 enterprise technology leaders have shifted their focus from AI adoption to AI accountability. A written rule is what accountability is measured against.
In the source’s own words
- Greyhound Research, November 5, 2026 front door notefinds 84 percent have shifted their focus from AI adoption to AI accountabilitySource
2How does this square with Microsoft’s message on trust?
Microsoft agrees on transparency. The keynote said that nested black boxes are unacceptable, announced an AI Transparency Initiative with 40 organizations, and described an end-to-end trust platform. Those are welcome commitments. They describe how Microsoft will show what its systems do. They do not write the customer’s rules for what its agents may do, which is the gap the series finds.
Why the trust quotes matter. Calling nested black boxes unacceptable is a strong line for a platform vendor to use on its own stage, and it gives buyers a standard to quote back in negotiation. Transparency of the platform and authority of the agent are different layers, though. The first lets a company see what happened. The second decides what may happen. Microsoft’s commitments address the first, and the series’ finding concerns the second, so the two positions fit together rather than conflict.
In the source’s own words
- The Official Microsoft Blog, Nadella keynoteThese nested black boxes are simply unacceptable.Source
- The Official Microsoft Blog, Nadella keynoteAlready, 40 organizations are working with us on defining and applying these standards.Source
- The Official Microsoft Blog, Nadella keynoteWe are building an end-to-end trust platform.Source
Part 2The Through-Line
The Series: Eight Notes, Eight Controls
Greyhound SpotlightThe series names eight controls, and only one of them, the spending policy, belongs to finance. The rest fall to security, data, legal, procurement, and the CIO.
The series has seven numbered notes and one extra on confidential computing. Each follows the same plan: what Microsoft announced, what it means for a buyer, the risks, the strongest case against the note’s view, a ten-move Greyhound CXO Runbook, and six questions for Microsoft. Each title follows one pattern. It names the change, and the control that change creates.
The series is built around controls rather than products for a reason. Microsoft’s product names and bundles change often, and features move between suites and licenses, while the questions a buyer must answer about authority, spend, and exit stay stable. Reading by control is likely to outlast this year’s packaging. The larger change is to the vendor relationship. Most enterprises run Microsoft through one account owner and one renewal. Once finance, security, legal, and data protection each own a control, each will bring its own questions to Microsoft, and a company that does not gather them in one place will negotiate in several voices at once, which tends to favor the seller.
Table 2 maps the eight notes, with the control each names, what the buyer writes, and the readers each is written for.
Greyhound Research · Table 2 · The Series
Eight notes, eight controls, one direction
Greyhound Research’s Microsoft Ignite 2026 series, note by note, with the control each names and the readers it is written for.
Every control needs a named owner. Read first the note whose control your role will own.
| Note | Research note | The control, and what the buyer writes | Written mainly for |
|---|---|---|---|
| 1 | Microsoft Copilot As The OS For Work | Frontier enrollmentWhich users, holding which data, may use the Frontier side of the new front door | CIOs, change and adoption leads, data protection officers |
| 2 | Microsoft Agent Economics | The spending policyWho may spend Copilot Credits, how much, on which services, and what happens at the limit | CFOs, FinOps leads, procurement |
| 3 | Microsoft Agent Identity | EnforcementWhere each agent’s credentials are issued, which is where it can be stopped | CISOs, identity leads, general counsel |
| 4 | Microsoft Azure Agent Infrastructure | AuthorityWhat each agent may read, change, and spend, and proof that it stayed inside those limits | Platform engineering, CISOs, regulated CIOs |
| CC | Microsoft Confidential Computing | AttestationProof of what ran in the seal, checked by someone the buyer trusts and refreshed when the agent changes | CISOs, data protection officers, chief risk officers |
| 5 | Microsoft Perception | The auditA regular, independent check of the alerts security agents closed on their own | CISOs, SOC leads, procurement |
| 6 | Microsoft MAI Models | Model choiceWho picks the model for each task, on which terms, and with whose data | CTOs, AI platform leads, general counsel |
| 7 | Microsoft Dynamics 365 Activate | The exit planA scoped, priced, tested plan to leave, used at every renewal | CIOs, business applications leads, CFOs |
Table 2 maps the eight companion notes of the Microsoft Ignite 2026 series, each with its control and main readers. Note 1, Copilot as the OS for work: Frontier enrollment, for CIOs and change leads. Note 2, agent economics: the spending policy, for CFOs and FinOps leads. Note 3, agent identity: enforcement, for CISOs and identity leads. Note 4, Azure agent infrastructure: authority, for platform engineering and CISOs. The confidential computing note, a series extra: attestation, for CISOs and data protection officers. Note 5, Perception: the audit, for CISOs and SOC leads. Note 6, MAI models: model choice, for CTOs and general counsel. Note 7, Dynamics 365 Activate: the exit plan, for CIOs and business applications leads.
The notes can be read in any order, though they build on each other. The note on Copilot as the front door and the one on agent economics cover what every user and every budget will meet first. The notes on agent identity, Azure agent infrastructure, and confidential computing follow an agent from its name to its machine and its sealed data. The notes on Perception, MAI models, and Dynamics 365 Activate cover the security agents, Microsoft’s own models, and the way out of rivals’ software.
Greyhound Standpoint:
Greyhound Research’s reading is that the most practical lesson of the series concerns ownership. The eight controls belong to at least six different functions, and no single team can write the rules for a workforce of agents on its own. Enrollment touches change management, the spending policy belongs with finance, enforcement and the audit sit with security, the data map with data protection, and model choice and the exit plan with procurement and legal. When ownership stays with whoever bought the license, the controls outside that team’s expertise tend to go unwritten, less because anyone decided against them than because no one was asked.
Consolidation cycles have taught this before. When enterprises standardized on large suites, the savings showed up in procurement while the governance work spread across functions that had not budgeted for it. Each note names its readers, so a leader can read the two or three that touch their remit. The assignment suggested here is a starting point, and a company with a central AI office may group the controls differently, which is reasonable. CIOs should assign each of the eight controls to a named executive at the next steering committee, set a date for a first draft, and record any control without an owner as an open risk.
The Deeper Read · About 1 minute
Select a question to open the full answer
1How do the notes relate to each other?
They agree by design. Where two notes meet, they reach the same verdict and say so. The note on agent identity and the note on Azure agent infrastructure both find AWS ahead today on written rules at the tool call. The notes on Perception and infrastructure both treat agents as insiders that need the narrow rights of a privileged account. The note on agent economics agrees with the infrastructure and confidential computing notes on the same week’s news.
Each keeps its own evidence. Every companion note sets out its own sources, evidence register, and limits. This overview summarizes their positions and does not restate their evidence. A reader checking a claim about a single product should go to the note that covers it.
Why agreement across notes matters. Where two notes reach the same verdict from different evidence, as with AWS’s lead on written rules at the tool call, the finding is sturdier than either note alone. Where a note records that Microsoft’s own pages disagree, the disagreement is information too. It usually marks a feature whose status or terms are still settling, and those are the features a buyer should hold at the pilot gate.
Part 3The Eight Notes In Brief
The Front Door And The Bill: Copilot And Agent Economics
Greyhound SpotlightThe front door moves for every licensed user this winter. The agents behind it stay under Frontier terms that leave out HIPAA coverage and three ISO standards.
The first thing most employees will notice from Ignite is a new starting page. Home, in the Copilot app, is where work now begins, with Word, Excel, and PowerPoint inside Copilot. Both are generally available. The agents that do work on a person’s behalf, Autopilot, Workspaces, and Today, are in the Frontier program. The companion note on Copilot as the front door finds that Frontier terms do not include HIPAA coverage and do not meet ISO 27001, 27002, or 27018. It also finds that Microsoft’s launch post says Autopilot leaves a clear audit trail while its documentation says Autopilot activity is not a guarantee of a complete audit record.
The second thing most companies will notice is the bill. Microsoft kept Microsoft 365 Copilot at USD 30 per user a month, Agent 365 at USD 15, and E7 at USD 99. Beside those seat prices it widened Copilot Credits, which now also pay for Windows 365 for Agents, a billing option in Fabric, and the Today Feed. The companion note on agent economics finds that credits are one currency but not yet one bill, because services are set up in three different consoles. It finds that a spending limit reserves nothing, stops the user’s work for the rest of the month when reached, and by default covers every new metered service. For new Copilot Business subscriptions bought through Microsoft’s partners, Microsoft plans to switch usage-based billing on by default between January and March 2027, moved from December 1, 2026, with a default limit of 4,000 credits per user a month that administrators can adjust.
Together the two notes describe the employee’s side of Ignite. One changes where people work, the other how the work is paid for. Both put a control in the buyer’s hands. The first is an enrollment list for the Frontier side of the front door. The second is a spending policy that says who may spend, how much, on which services, and what happens when the money runs out.
The shift beneath both notes is from a fixed bill to a variable one. For most of the history of Office licensing, the seat price was close to the whole cost, and finance could forecast it from headcount. A second line now moves with use, and it sits behind the very front door Microsoft is designing to be used more. Greyhound Research expects the first surprises to come through partner-billed Copilot Business subscriptions once usage billing switches on by default in early 2027. The effect most coverage misses is that adoption and cost now pull in the same direction. A rollout that succeeds in getting people to use agents will, by design, raise the bill.
Greyhound Standpoint:
Greyhound Research holds that the front door and the bill should be governed as one program, because the same employees will meet both in the same months. When the front door moves, Frontier enrollment becomes the control. When work is billed by the task, the spending policy becomes the control. Each user enrolled in a Frontier agent feature is also a new point of spend, and a spending limit, as the economics note finds, reserves nothing and stops the user’s work for the rest of the month once reached. Enrollment without a policy invites unplanned cost, and a policy without enrollment discipline invites stalled work.
Interface changes and pricing changes have usually been run by different teams, and the friction between them has tended to surface at the help desk and in the budget review. Microsoft deserves credit for holding seat prices steady and for publishing more about the cost of a task than its rivals. The limits are that several credit rates are unpublished and that Frontier terms do not yet include HIPAA coverage or three ISO standards, so the right scope for enrollment may widen once release terms appear. Change leaders should retrain users by role this quarter and enroll only named pilot groups. Finance leaders should switch off automatic coverage of new metered services and write a spending policy for each business unit before the next renewal.
The Deeper Read · About 2 minutes
Select a question to open the full answer
1What does the note on Copilot as the OS for work find?
The doorway now, the agents slowly. The companion note finds that Home and Office in Copilot change where every user starts work, and that Microsoft’s rollout notices carry no training guidance. Its ten moves begin with retraining by role, enrolling named groups in Frontier, and keeping regulated data out of Frontier features. Its advice is to move the doorway now and enroll the agents slowly. It finds Microsoft ahead of Google on released features and on written terms.
Signals. Release dates for Autopilot, Workspaces, and Today, HIPAA and ISO coverage at release, a complete Autopilot audit record, and a stated retention period for Autopilot’s memory would each strengthen the case for wider enrollment.
What the gap between the launch post and the documentation tells a buyer. The launch post says Autopilot leaves a clear audit trail, and the documentation says its activity is not a guarantee of a complete audit record. Both can be true. A launch post describes intent, and documentation describes what Microsoft commits to. For audit and legal purposes the documentation is the page that counts, so a company relying on Autopilot for compliance should treat its trail as useful evidence, not as a complete record.
2What does the note on agent economics find?
The policy, not the price list. The companion note finds that predictability has moved from Microsoft’s price list to the spending policy the customer writes. It finds Microsoft ahead of its rivals in publishing what a task costs, level with them in pairing seats with a meter, and behind Google on what happens at the cap, where Google pauses an agent’s calls and Microsoft stops the user’s access for the month. It also finds that Microsoft now meters the spend and scores the value, so the business should keep its own measure of what the work was worth.
How the view on credits has moved. Greyhound Research said in November 2025 that new units such as credits were “deliberately designed to obscure value exchange.” The economics note finds that Copilot Credits answer much of that charge, because Microsoft now publishes more of what a task costs than its rivals do, and keeps the earlier caution where rates are still unwritten.
What is missing. Several credit rates are unpublished, including those for the Today Feed, Fabric, and Windows 365 for Agents. Its advice is to buy prepaid credits from measured use and to negotiate the terms for credits at renewal as hard as the price of seats. Microsoft’s Source summary says Business Analytics shows leaders what their AI is costing.
In the source’s own words
- Microsoft Source, November 17, 2026Agent 365 Business Analytics gives leaders a clear view of what their AI is costingSource
Part 3The Eight Notes In Brief
The Name, The Machine, And The Seal: Identity, Infrastructure, And Confidential Computing
Greyhound SpotlightMicrosoft can now give a Salesforce or AWS agent a Microsoft name. Its own documentation describes that link as read-only, so stopping the agent stays with the platform that runs it.
Three companion notes follow an agent through the infrastructure that runs it. The first is its name. Microsoft’s connected platforms feature copies agents from Salesforce, AWS, Google, Anthropic, and other vendors into Agent 365 and gives each an Entra Agent ID. The companion note on agent identity finds that Microsoft’s documentation calls the link to rivals’ platforms read-only. Agent 365 can name and watch a rival’s agent. It cannot stop one, so enforcement stays wherever each agent’s credentials are issued.
The second is its machine. The note on Azure agent infrastructure finds that Microsoft gave agents sealed-off machines, network rules that refuse by default, and a cache of work already done, and brought Foundry onto customers’ own clusters. Running agents is level with AWS and Google, AWS is ahead on written rules for each tool call, and Microsoft is ahead on the identity, security, and Microsoft 365 estate around the agent. The third is its data. The note on confidential computing finds the seal keeps the cloud operator out but not the customer’s own agent, which is the one reading the data.
The detail that matters most in these three notes is the read-only link. For human staff, the directory has long been the master record: disable an account there and access ends almost everywhere. For agents on rivals’ platforms, Microsoft’s directory is a mirror of a record held elsewhere. With AWS also accepting outside agents in its registry, many large buyers are likely to hold more than one list of the same agents. The second-order effect falls on incident response. A responder’s first move will be to disable the agent in Entra, and for a rival’s agent that will not stop it, so the incident runbook must say where to go next and who answers there.
Exhibit 4 follows an agent through five steps, from being named to being stopped, and grades how much of each Microsoft has written.
Greyhound Research · Exhibit 4 · The path of an agent’s authority
Microsoft has written the name and the machine, not the limit
Five steps from giving an agent a name to being able to stop it, read against what Microsoft had documented by November 19, 2026.
The unwritten steps decide what an agent can do. That is where the buyer’s own rule has to go first.
Exhibit 4 sets out five steps of an agent’s authority as a numbered list, and shows Microsoft has written the first two and not the last two. Step one, name: written, as Entra Agent ID reaches rivals’ agents through connected platforms, though the link to rivals’ platforms is read-only. Step two, house: written, with Container Apps Sandboxes generally available, while the AKS sandbox extension had no documentation. Step three, seal: partly written, with the Confidential Clean Room in limited preview, the other confidential computing items still in development, and no statement on agent memory. Step four, limit: not written, as Agent Grid had no documentation. Step five, stop: not written for rivals’ platforms, though Microsoft enforces on Windows devices. The grades are Greyhound Research analysis.
Microsoft has written the first two steps and part of the third. The limit at each tool call and the power to stop a rival’s agent are the parts not yet written, and they are the parts that decide what an agent can do. The three notes reach one finding from three directions. Isolation and sealing keep others out. Neither decides what the agent itself may do with the access it holds.
Greyhound Standpoint:
Greyhound Research’s view is that Microsoft has built the most complete directory and surrounding estate for agents of any vendor read for the series, and has left the layer that decides what an agent may do unwritten. When rivals’ agents get a Microsoft name, enforcement becomes the control. When agents get their own machines, authority becomes the control. When agents work inside the seal, attestation becomes the control. A name lets a company find an agent, isolation keeps other workloads out, and the seal keeps the cloud operator out. None of them limits what the agent does with the access it holds, which is where the exposure for data and money sits.
Identity programs have followed this arc before. Directories came first, and the discipline of limiting what powerful accounts may do followed later, at greater cost. Credit is due on several counts. Running agents on Azure is level with AWS and Google, and Microsoft leads on the identity, security, and Microsoft 365 estate around the agent, while AWS is ahead today on written rules at each tool call. A documented rule at the tool call would change this view. Before the next architecture review, CISOs should register every agent as a privileged account with a named sponsor and map where its credentials are issued. Regulated buyers should commit sensitive data only once the proof, its checker, and the agent’s memory are documented.
The Deeper Read · About 3 minutes
Select a question to open the full answer
1What does the note on agent identity find?
A directory entry is not a control. The companion note finds that Microsoft names eleven connected platforms while its documentation lists seven, syncs only when an administrator asks, and does not promise to remove agents deleted at the source. On employees’ computers Microsoft does enforce, through Defender, Intune, and Windows containers, with a warning that a block can arrive after a tool has run. Its advice is to use Agent 365 as the inventory of record and keep the power to stop each agent with the platform that issues its credentials. Microsoft is not alone in bidding for the cross-vendor list: AWS’s agent registry, generally available since August 31, 2026, also accepts agents running outside AWS. The note finds Microsoft ahead on reach, AWS ahead on rules at each tool call, and no vendor documenting enforcement on a rival’s platform.
Microsoft’s own description agrees. Ray Smith, Microsoft’s corporate vice president for the Agent 365 product, described Agent 365 in a partner’s launch release as the observability platform for AI, helping organizations turn shared visibility into management, governance, and security decisions. Security partners announced the same week that they would add runtime evidence and enforcement alongside it. Observability is a real and useful layer. The decisions it informs still have to be written by the customer.
Why the word observability matters. Ray Smith’s choice of word is precise, and it matches the documentation’s read-only label. An observability platform shows what agents do. By that description, it does not decide what they may do. Buyers should take Microsoft at its word here. The security partners’ announcements the same week point the same way: enforcement is expected to come from a mix of Microsoft and other vendors, which means the buyer will be the one assembling it.
2What does the note on Azure agent infrastructure find?
Ready to pilot. The companion note finds that Ignite made running agents on Azure ready to pilot. Almost all of it is in preview, with Cobalt 200 and Container Apps Sandboxes generally available. It finds that Context Cache saves money only where agents reread the same long text and is a new store of data to govern, since it keeps processed text for 1 to 30 days counted from last use, so a cache in steady use does not expire. Its section on confidential computing agrees with the companion note: one limited preview and two items still in development. It places lock-in in an agent’s identity, memory, and records, not in the model it calls, and sets out ten moves.
3What does the note on confidential computing find?
The proof matters more than the lock. The companion note finds three announced items without prices or release dates: one limited preview, the clean room, which restates a June preview, and two items still in development. Microsoft has not said whether agent memory and cached context sit inside the seal, and its Ignite demonstration checked the hardware’s proof with a local verifier, with remote attestation described as an architectural direction, while Google already offers a verifier it does not own. Its advice is to pilot on what is generally available and commit regulated data only when the proof, its checker, and the agent’s memory are documented.
In the source’s own words
- Vikas Bhatia, Microsoft, Azure confidential computing at Ignite 2026Remote attestation is an architectural direction, while the Ignite demonstration uses a local verifier.Source
Part 3The Eight Notes In Brief
The Defender, The Model, And The Exit: Perception, MAI, And Activate
Greyhound SpotlightIn three areas Microsoft now plays two roles at once. It supplies the security agents and their price, sells its own models beside its partners’, and sells both the way out of Salesforce and a way to stay.
Microsoft Perception, a team of AI security agents inside Microsoft Defender, became generally available, though three of Microsoft’s own pages still call it a preview. The companion note on Perception finds that its Triage Agent closes alerts it judges to be false alarms without asking anyone first, and Microsoft says it covers about 80% of alerts in a typical Defender queue. The note reads that figure as coverage, not accuracy. Its price has a meter and a calculator but no published rate card, and Microsoft’s own pages give three versions of the Sentinel offer for E5 and E7 customers.
Microsoft also put its own MAI models on sale beside the OpenAI and Anthropic models it resells. The companion note on MAI models finds that MAI grows first inside Microsoft’s own products, starting with the Copilot model picker. It is a credible option for everyday coding, routing, and classification, likely to become a firm price anchor once list prices are published and others can test it, with cost claims measured at a promotional price and contract terms still catching up with fast product progress. Microsoft’s copyright protection page does not yet name MAI models. MAI Licensed Weights, which would let a company run a tuned MAI model in its own Azure account, has no price or date. The note on Dynamics 365 Activate finds Microsoft selling the exit from Salesforce and the stay with it in the same week, with both roads ending at Copilot.
In each of the three areas, the buyer needs a record Microsoft does not mark. For Perception, it is a regular check of what the agents closed. For MAI, it is the company’s own test of a model on its own code, and its own rule for which model serves which task. For Activate, it is a priced, tested plan for leaving that the buyer owns whether or not it moves.
The new element is how far Microsoft now competes with the companies it works beside. It resells OpenAI and Anthropic models while selling its own, and in the same week it sells a road out of Salesforce and a way to stay with it. Rivals are likely to answer at renewal time, which gives buyers more room to negotiate than they had a year ago. The second-order effect most coverage misses sits in the security team. When agents close alerts without asking, analysts review fewer cases, and over time the skill to judge a closure can thin. A regular sample of agent decisions keeps that skill in practice as well as checking the agents.
Greyhound Standpoint:
Greyhound Research holds that when a vendor plays two roles at once, the buyer’s own record becomes the control. When agents close the alerts, the audit becomes the control. When the platform owner ships its own models, model choice becomes the control. When leaving gets cheaper, the exit plan becomes the control. Two roles are not a problem in themselves, but the party reporting on the work also has an interest in the result, so an independent measure belongs beside the vendor’s own.
Each area also shows real progress. Perception brings red, blue, and green team agents into one console. MAI is a credible option for everyday coding and routing, and is likely to become a firm price anchor once list prices are published, the models are sold beyond Microsoft’s own tools, and others can test them. Earlier consolidation cycles taught the discipline this calls for: use the vendor’s tooling, and check its claims against the company’s own data. The limits are that Perception has no published rate card, MAI’s cost claims rest on a promotional price, and Activate’s speed figures are partner forecasts and company claims. SOC leads should audit a weekly sample of agent-closed alerts. AI platform leads should pilot MAI on the company’s own work and pin the model for each task. CIOs should commission Activate’s discovery before the next Salesforce or SAP renewal.
The Deeper Read · About 3 minutes
Select a question to open the full answer
1What does the note on Perception find?
Let the agents work, and check their work. The companion note records that the Book of News calls Perception generally available while three of Microsoft’s own pages still say preview: the older product page calls it a preview, and two documentation pages an invitation-only limited preview. Its status therefore rests on the Ignite announcements. It finds Microsoft ahead on breadth, with red, blue, and green team agents in one console, and on bundling a cheaper SIEM for E5 and E7. The agents that defend the company are insiders too, and need narrow rights and reviews. It finds Perception priced separately from Security Copilot, whose capacity units many E5 customers already hold. Among rivals, CrowdStrike lists approval gates, under which an agent pauses for an analyst before a sensitive tool call, as available today, while Perception’s Triage Agent has no approval step before a closure.
Why the status label matters. A status label is more than a formality. It decides which service terms, support commitments, and compliance coverage apply. When the Book of News says generally available and three Microsoft pages still say preview, a buyer cannot tell from the pages alone which terms govern its use. The practical answer is to ask Microsoft to confirm the status in writing for the buyer’s own contract, and to file that confirmation with the pilot decision.
In the source’s own words
- Microsoft Source, November 17, 2026Microsoft Perception, now generally available, brings together specialized agents to find, defend and remediate vulnerabilitiesSource
2What does the note on MAI models find?
Supplier, reseller, router, and rival. The companion note finds that MAI Code 2 is first an expansion inside Microsoft’s own products, with a wider contest in the open market likely to follow, and that by Microsoft’s own account it is not a replacement for OpenAI or Anthropic models on the hardest work. Microsoft disclosed its test conditions more fully than most launches do. Its headline cost ratios use a promotional price with no list price yet published, an earlier MAI coding model was replaced about eleven weeks after reaching business customers, and the smaller coding model was trained in part on prompts from individual GitHub Copilot users who had not opted out. Microsoft corrected its own blind preference figure on November 18, and both of its pages now give 64%, an evaluation that has not been independently reproduced. Licensed Weights also has a precursor: on October 15, 2026, Microsoft AI said its MAI-Base-2 base model would let it license model weights to trusted partners through Foundry. Its advice is to pilot MAI on the company’s own work, budget on the cost of each change that ships, keep the right to pick and pin the model, and explore licensed weights with Microsoft while price, export, indemnity, and update terms are agreed.
In the source’s own words
- The Official Microsoft Blog, Nadella keynoteWe are announcing MAI Licensed WeightsSource
3What does the note on Dynamics 365 Activate find?
The discovery report is the value. The companion note finds no published price for Activate, a SAP tool in private preview while SAP’s mainstream maintenance for ECC ends at the end of 2027, and speed figures that are partner forecasts and company claims without a method. Microsoft’s own pages disagree on the SAP tool’s status, and a Microsoft partner page still describes the Salesforce tool as a limited public preview. It records that Microsoft names Comcast and Fusion5 among Activate’s customers, though Fusion5 is a Microsoft partner, and that Comcast’s Sathish Arumugam credits the discovery step with more informed planning, which speaks to the assessment rather than a completed move. Accenture, the design and implementation partner for Copilotforce, calls itself customer zero. It finds that Copilotforce keeps Salesforce’s own permissions and that its licensing is not written, so a seller could be paid for twice. Its advice is to run the discovery, price the exit, and keep press-release timelines out of contracts.
Part 4The Risks And The Counter-Argument
Risks To Watch: Where The Series’ Controls Can Fall Short
Greyhound SpotlightNone of the five risks needs Microsoft to get anything wrong. Each starts when a buyer’s own rule is missing.
Greyhound Research sees five risks that run across the series. Each appears in more than one companion note, and each shows first inside the buyer’s own organization, in a pilot list, a policy, a register, or a board report. That makes them cheap to catch early and slow to unwind later.
These risks differ from the ones most technology risk registers were built for. Registers have tended to list vendor failures, such as an outage, a breach, or a missed service level, each with a contract remedy attached. The five here are gaps in the buyer’s own records, and no service credit covers them. Greyhound Research expects the first signs to appear in expense reviews and internal audit findings rather than in security incidents. The second-order effect is on accountability. When a loss traces back to a rule nobody wrote, the company has no supplier to hold to account, and the board will ask who inside the firm owned the gap.
Table 3 sets each risk beside its early warning and its owner. Every early warning can be checked without help from Microsoft.
Greyhound Research · Table 3 · Risk register for the series
Five risks, and each starts in the buyer’s own paperwork
Five risks that run across the Microsoft Ignite 2026 series, each with its early warning and its owner.
None of the five needs a Microsoft error. Each is cheapest to catch in the buyer’s own records this quarter.
| Risk | Early warning in the buyer’s own estate | Owner |
|---|---|---|
| Preview drifts into productionFrontier and preview features reach real work before their terms change. | A pilot group that grows without a new decision. | CIO |
| A meter without a policyCredits and security agents draw on a shared pool with no owner per unit. | A new metered service that joined existing policies by default. | CFO and FinOps lead |
| A name without a stopAgents are listed in Entra but can only be halted on their own platform. | An agent inventory with no column for where credentials are issued. | CISO |
| Unwritten data storesAgent memory, cached context, and Frontier features hold data outside written terms. | A data register with no entry for agent memory or caches. | Data protection officer |
| The seller reports on its own workMicrosoft meters the spend, scores the value, and grades its own agents. | A board report built only from Microsoft’s dashboards. | Chief risk officer |
Table 3 lists five risks that run across the Microsoft Ignite 2026 series, each with an early warning in the buyer’s own estate and an owner. Preview drifts into production, warned by a pilot group that grows without a new decision, owned by the CIO. A meter without a policy, warned by a new metered service joining existing policies by default, owned by the CFO and FinOps lead. A name without a stop, warned by an agent inventory with no column for where credentials are issued, owned by the CISO. Unwritten data stores, warned by a data register with no entry for agent memory or caches, owned by the data protection officer. The seller reports on its own work, warned by a board report built only from Microsoft’s dashboards, owned by the chief risk officer. The risks are Greyhound Research analysis.
The five share a cause: a belief that a released or bundled tool comes with the rule it enforces. Preview drift and unwritten data stores come from treating test terms as contract terms. A meter without a policy and a name without a stop come from treating a tool as a control. The last risk is the newest. Microsoft now meters the spend, scores the value, grades its own security agents, and sells its own models, so a board that reads only Microsoft’s dashboards is reading the seller’s own account of its work.
Greyhound Standpoint:
In Greyhound Research’s reading, the five risks share one cause and one remedy. Each begins when a company treats a released or bundled tool as if it came with the rule it enforces, and each is cheaper to prevent with a written rule than to repair after an incident. None requires Microsoft to make an error. A pilot group that grows without a new decision, a metered service that joins a budget by default, or a register with no column for where an agent can be stopped is a gap in the buyer’s own paperwork, and the early warning sits there too.
Operational risk has tended to build this way in earlier technology shifts. Many of the incidents that reach boards begin as small, unowned exceptions that persist for months before anyone measures them. None of the five reflects a fault in Microsoft’s products, and none should be read as a forecast that harm will occur. General availability terms for the agent features and a published rate card would take two of the risks partly off the buyer’s desk, and would ease this view. Chief risk officers should test all five warnings this quarter, name an owner for each, and report the result at the next board risk review.
Part 4The Risks And The Counter-Argument
The Strongest Objection: Microsoft Will Write The Controls Itself
Greyhound SpotlightThe strongest objection to this note is that Microsoft already sells the controls, in Agent 365, Entra, Purview, Defender, and E7. That part is right.
The objection deserves its full weight. Microsoft ships more control tooling for agents than any rival read for the series. It leads on the reach of its directory, on the breadth of its security agents, and on publishing what a task costs. E7 bundles Microsoft 365 E5, Copilot, and Agent 365 in one license. On this view, a company that buys the suite and switches on its controls has done the work, and a written rule from the buyer adds paperwork, not protection.
The objection has force because the controls are already paid for. A company on E7 holds Entra Suite and advanced Defender, Intune, and Purview capabilities in the same license, and switching them on can feel like the end of the work. Greyhound Research expects Microsoft to narrow part of the gap over time with templates and defaults, as vendors usually do once a product matures. The second-order effect is easy to miss. When a company’s controls live inside a license tier, its control framework becomes part of the renewal. Moving down a tier, or moving a workload elsewhere, then removes controls as well as features, and that belongs in the renewal analysis.
Exhibit 5 tests each part of the objection against the evidence in the series. Two parts stand, and two do not hold.
Greyhound Research · Exhibit 5 · The objection held to the evidence
Two parts of the objection stand, and two do not hold
The fair counter-case to this research note, that Microsoft will write the controls itself, part by part against the evidence in the series.
The objection holds for the tools, not for the rules. Switch the tools on and own the rules they enforce.
Exhibit 5 holds four parts of the strongest objection to this research note against the evidence, as four cards, and shows two stand and two do not hold. Microsoft ships more control tooling than its rivals: stands, as it leads on directory reach, bundled security agents, and publishing what a task costs. One suite simplifies buying the controls: stands, as E7 bundles Microsoft 365 E5, Copilot, and Agent 365 at USD 99 per user. The tools are the control: does not hold, because a spending limit is a ceiling, the link to rivals’ agents is read-only, and Autopilot’s audit record is not guaranteed complete. Buyers can wait for general availability: does not hold on timing, because the new front door reaches every licensed user this winter. The verdicts are Greyhound Research analysis.
The tooling and the bundle are real, and buyers should use both. The claim that the tools are the control does not hold on the series’ own evidence. A spending limit is a ceiling that stops work, not a budget. The link to rivals’ agents is read-only. Autopilot’s audit record is not guaranteed complete. The claim that buyers can wait does not hold on timing, because the doorway reaches every user before the agent terms are written.
Greyhound Standpoint:
Greyhound Research accepts the objection for the tools and does not accept it for the rules. Microsoft ships the strongest set of agent controls of any vendor read for this series, across Agent 365, Entra, Purview, and Defender, and E7 brings many of them into one license. Buying them and switching them on is sensible. The objection weakens at one step. A control enforces a rule, and the rule has to come from someone who knows the business. A spending limit cannot know which workflow must keep running at month end, and a directory cannot know which rival’s agent should be stopped first.
Seasoned buyers have heard versions of this argument in earlier consolidation cycles, from integrated enterprise application suites to bundled security platforms. Each time the bundle simplified purchasing, and the policy work stayed with the customer. That is the normal division of labor between the vendor, who builds the mechanism, and the customer, who decides its use. The view would shift if Microsoft published a documented rule at the tool call and a per-platform list of what it can enforce. Until then, the two together are the control, and either alone is not. Leaders should switch Microsoft’s tools on at the next pilot gate and, in the same decision, approve the written limits those tools will apply, each with a named owner.
The Deeper Read · About 1 minute
Select a question to open the full answer
1Why do the tools not settle the control?
Three findings from three notes. The note on agent economics finds that a spending limit reserves nothing and stops the user’s work for the rest of the month when reached. The note on agent identity finds the link to rivals’ platforms read-only. The note on Copilot as the front door finds that Autopilot activity is not a guarantee of a complete audit record. In each case Microsoft’s tool works as designed. The rule it needs is the buyer’s.
The bundle is real. Microsoft says E7 unifies Microsoft 365 E5, Microsoft 365 Copilot, and Agent 365, and adds Entra Suite and advanced Defender, Intune, and Purview capabilities. That simplifies buying the tools. It does not decide who in the company may use them for what.
What “unifies” tells a buyer. Microsoft’s word for E7 is accurate for buying. One contract and one price replace several. Unifying the purchase does not unify the decisions. Each tool in the bundle still asks a different team a different question, from Purview’s data labels to Defender’s response rules, and the bundle makes those questions arrive together rather than one product at a time.
In the source’s own words
- The Official Microsoft Blog, March 9, 2026Microsoft 365 E7 unifies Microsoft 365 E5, Microsoft 365 Copilot and Agent 365 into a single solutionSource
Part 5What To Do, By Role
The Greyhound CXO Runbook: Ten Moves Across The Series
Greyhound SpotlightEvery move produces a document the buyer owns, from an enrollment list, a spending policy, and an agent register to a status record, a renewal analysis, and a rehearsed fallback.
Ten moves follow from the series as a whole. Each companion note sets out its own ten moves for its own area. These ten sit above them, at least one for each leader who will own part of a company’s agent workforce. None waits for Microsoft, because each uses controls the buyer already holds.
Greyhound Research has kept the moves to documents on purpose. Earlier platform shifts often produced programs and committees whose output was hard to inspect. A list, a policy, a register, or a clause can be read, dated, and checked, which makes each move easy to audit and hard to defer quietly. The first versions will be imperfect and should be revised as Microsoft publishes terms. The benefit most coverage misses is portability. None of the ten documents is specific to Microsoft. A spending policy, an agent register, or an exit clause written now will serve as well for agents from AWS, Google, or Salesforce, so the work reduces dependence on any one vendor rather than adding to it.
Table 4 sets the ten moves beside the assumption each corrects, its evidence, and its owner. Two of the ten finish only when Microsoft publishes its terms.
Greyhound Research · Table 4 · The Greyhound CXO Runbook by owner
Ten moves across the series, and none waits on Microsoft to start
Ten series-level moves for the leaders who will own a part of Microsoft’s agent estate, with the assumption each corrects, its evidence, and its owner.
Treat all ten as this quarter’s work, not next year’s. Only two need Microsoft’s published terms to finish.
Select a move to open its full text in The Deeper Read below
| Move | The assumption it corrects | Evidence anchor | The decision sits with |
|---|---|---|---|
| 1/Sign on general availability terms, and enroll named groups only | A keynote status is a contract status | Exhibit 1, Table 1 | CIO, with procurement |
| 2/Write the spending policy before the meter runs | A price list makes AI spend predictable | Table 3 | CFO, with the FinOps lead |
| 3/Register every agent as a privileged account, and map where it can be stopped | A directory entry is a control | Exhibit 4 | CISO, with identity leads |
| 4/Map where agent data rests, and keep regulated data in written places | A sealed or licensed service covers everything the agent touches | Exhibit 4, Table 3 | Data protection officer |
| 5/Write model choice and the exit into every renewal | Model and platform are settled at signing | Exhibit 3 | Procurement, with general counsel |
| 6/Keep an audit of agent work that Microsoft does not mark | The vendor’s dashboard is the record | Exhibit 5, Table 3 | Chief risk officer, with the CIO |
| 7/Retrain users by role before the front door moves | A new starting page needs no change program | Table 1, Exhibit 5 | Change and adoption leads |
| 8/Approve features, not product names, and get each status in writing | One product name carries one status | Table 1 | General counsel |
| 9/Price the controls inside the license tier before any renewal | A license tier is a price decision alone | Exhibit 5 | CFO, with the CISO |
| 10/Rehearse a fallback for the agent work that cannot stop | The suite’s resilience covers the agents inside it | Exhibit 2 | Chief risk officer |
Table 4 sets out the ten moves of the series-level Greyhound CXO Runbook, each with the assumption it corrects, the exhibit or table it rests on, and the executive with whom the decision sits, and shows that none waits on Microsoft to start while two complete only when Microsoft publishes terms. The moves are: sign on general availability terms and enroll named groups only, write the spending policy before the meter runs, register every agent as a privileged account and map where it can be stopped, map where agent data rests, write model choice and the exit into every renewal, keep an audit of agent work that Microsoft does not mark, retrain users by role before the front door moves, approve features rather than product names and get each status in writing, price the controls inside the license tier before any renewal, and rehearse a fallback for the agent work that cannot stop. The moves are Greyhound Research analysis.
The two that finish later, the first move on signing terms and the fifth on renewals, wait on terms Microsoft has not yet published. The questions that follow ask for those terms, and each maps to a move a buyer can make meanwhile. The other eight can be finished this quarter with documents and tools the buyer already holds. Each move points to the companion notes whose own Runbooks carry the detail.
The Deeper Read · About 5 minutes
Select a move to open its full text
1CIOs: sign on general availability terms, and enroll named groups only.
Exhibit 1 and Table 1 show the released products beside agent features still on preview terms. This move corrects the belief that a keynote status is a contract status. Its action is to sign production commitments only on general availability terms, to enroll named pilot groups, not all users, in Frontier and previews, and to record each enrollment as a decision with an owner. The companion notes on Copilot as the front door and Azure agent infrastructure carry the detail. The decision sits with the CIO, together with procurement.
The order of the moves matters as much as the moves. Signing on released terms and enrolling named groups come first because every later document depends on knowing which agents are in use and under which terms. A spending policy, a register, or an audit sample written before that list exists will describe an estate that has already changed.
2CFOs: write the spending policy before the meter runs.
Table 3 shows a meter without a policy as a risk that starts by default. This move corrects the belief that a price list makes AI spend predictable. Its action is to switch off automatic coverage of new metered services, write one spending policy per business unit, budget the credit line as a range, and decide what happens at the limit for each critical workflow. The companion notes on agent economics and Perception carry the detail. The decision sits with the CFO, together with the FinOps lead.
3CISOs: register every agent as a privileged account, and map where it can be stopped.
Exhibit 4 shows Microsoft has written an agent’s name and machine but not the limit or the stop. This move corrects the belief that a directory entry is a control. Its action is to register every agent, security agents included, as a privileged account with a named sponsor, and to record for each one where its credentials are issued and who can halt it. The companion notes on agent identity, infrastructure, and Perception carry the detail. The decision sits with the CISO, together with identity leads.
4Data protection officers: map where agent data rests, and keep regulated data in written places.
Exhibit 4 and Table 3 show agent memory, cached context, and Frontier features holding data outside written terms. This move corrects the belief that a sealed or licensed service covers everything the agent touches. Its action is to list every place an agent’s data rests during a task, mark which places have written terms, and keep regulated data out of the rest. The companion notes on Copilot as the front door, Azure agent infrastructure, and confidential computing carry the detail. The decision sits with the data protection officer.
5Procurement: write model choice and the exit into every renewal.
Exhibit 3 shows price and terms unwritten in seven of the eight areas. This move corrects the belief that model and platform are settled at signing. Its action is to write the right to pick and pin a model, copyright cover for each route, and the export of agent identity, state, and records into every Microsoft renewal, and to bring a priced exit plan to every Salesforce and SAP renewal. The companion notes on MAI models and Dynamics 365 Activate carry the detail. The decision sits with procurement, together with general counsel.
6Chief risk officers: keep an audit of agent work that Microsoft does not mark.
Exhibit 5 and Table 3 show the seller metering the spend, scoring the value, and grading its own agents. This move corrects the belief that the vendor’s dashboard is the record. Its action is to keep a regular, independent sample of agent decisions, from closed alerts to Autopilot actions, and a value measure drawn from the business’s own records, and to report both to the board. The companion notes on Perception, agent economics, and Copilot as the front door carry the detail. The decision sits with the chief risk officer, together with the CIO.
7Change and adoption leads: retrain users by role before the front door moves.
Table 1 ranks the new front door first because it reaches every licensed user this winter, and Exhibit 5 shows it arriving before the agent terms. This move corrects the belief that a new starting page needs no change program. Microsoft’s rollout notices carry no training guidance, so the control is the buyer’s own training plan. Its action is to retrain users by role this quarter, starting with the named groups enrolled under the first move, to brief the help desk before Home arrives, and to tell each group which features are released and which are still on test terms. The companion note on Copilot as the front door carries the detail.
Training also shapes the bill. A rollout that gets people to use agents will, by design, raise it, so each training wave should be agreed with the owner of the spending policy in the second move. The decision sits with change and adoption leads, at each pilot gate.
8General counsel: approve features, not product names, and get each status in writing.
Table 1 shows status set by feature, with released and preview items under one product name. Of the 59 announcement headings in the Book of News, by Greyhound Research’s count, 28 mix the two. This move corrects the belief that one product name carries one status. Microsoft’s pages also differ. The Book of News calls Perception generally available, while three Microsoft pages still say preview. Its action is to list features, not product names, in the approvals, risk registers, and contract schedules it already keeps, to cite for each feature the most specific Microsoft page with the date it was read, and to ask Microsoft to confirm the status in writing for the buyer’s contract. The companion notes on Perception and Dynamics 365 Activate record where the pages disagree.
The confirmation is filed with the pilot decision and checked again at renewal, since statuses are likely to keep moving. The decision sits with general counsel, at each pilot gate and renewal.
9CFOs: price the controls inside the license tier before any renewal.
Exhibit 5 shows that one suite simplifies buying the controls, with E7 bundling Microsoft 365 E5, Copilot, and Agent 365 at USD 99 per user. This move corrects the belief that a license tier is a price decision alone. A company on E7 also holds Entra Suite and advanced Defender, Intune, and Purview capabilities in the same license, so its control framework becomes part of the renewal. Moving down a tier, or moving a workload elsewhere, then removes controls as well as features. Its action is to list, from the license records the buyer already holds, which agent controls depend on which tier, to name what would replace each one if the tier changed, and to put both in the renewal analysis beside the seat price. The companion note on agent economics carries the detail on the meter beside the seats.
The decision sits with the CFO, together with the CISO, before each Microsoft renewal.
10Chief risk officers: rehearse a fallback for the agent work that cannot stop.
Exhibit 2 ends with Ignite opening days after a Microsoft 365 outage, which Greyhound Research counts as the fourth multi-service Microsoft 365 incident of 2026. This move corrects the belief that the suite’s resilience covers the agents inside it. Enterprises have priced Microsoft 365 as a license rather than as a single point of failure, and agents raise the same question one level up. Work can also stop without an outage, because a spending limit, once reached, stops the user’s work for the rest of the month. Its action is to start from the company’s existing continuity plan, name the agent-supported processes that cannot stop, and rehearse for each a fallback that shares nothing with the suite, including its sign-in. A fallback that signs in through the suite shares its outage. The note on that outage carries the detail.
The decision sits with the chief risk officer, who reports the results of each rehearsal at the next board risk review.
Part 5What To Do, By Role
Questions For Microsoft: What Buyers Will Ask Next
Greyhound SpotlightFour of the six questions have no answer at all in what Microsoft has published, including what leaves with the customer.
Every finding in the series rests on what Microsoft has published, and Microsoft alone can close the gaps the series names. Each companion note asks six questions of its own, and the note on Azure agent infrastructure asks ten. The six below sit above them. Each gives what Microsoft has said so far, what the answer would settle, and the Runbook move it maps to. No move waits on them.
Questions like these carry more weight when they are put in writing and asked by several buyers at once. In past licensing transitions, the terms that many customers asked for tended to arrive sooner. Greyhound Research expects release terms and rates to be answered first, because Microsoft needs them to sell, and the export question last, because vendors seldom lead with the terms of leaving. The second-order effect is contractual. A written answer from an account team can be attached to an order form or a renewal schedule, which turns a briefing into a commitment the buyer can rely on. Asked only in a meeting, the same answer leaves no record.
Table 5 tracks the six questions against what Microsoft has said so far. Two are answered in part, and four not at all.
Greyhound Research · Table 5 · Questions for Microsoft, tracked
Two questions answered in part, four not at all
Six series-level questions only Microsoft can answer, against what it had stated by 1:20 a.m. UTC on November 19, 2026.
Ask in writing, and act meanwhile. A written answer can become a contract term, and no move waits on it.
Select a question to open its full text in The Deeper Read below
| Question | What Microsoft has said so far | The Runbook move it settles |
|---|---|---|
| Q1When will the agent features reach general availability, and on which terms? | Not stated December is named for some items. Autopilot, Workspaces, Today, Agent Grid, and Licensed Weights have no date. | Runbook move 1 |
| Q2Will every credit-using service, and Perception, have a published rate? | Stated in part Seat prices are held. Several credit rates and Perception’s rate card are not published. | Runbook move 2 |
| Q3For each platform, which controls are enforced and which are only observed? | Not stated Microsoft’s documentation calls the link to rivals’ platforms read-only. | Runbook move 3 |
| Q4Where do agent memory, cached context, and Frontier data rest, and under which terms? | Not stated Frontier terms leave out HIPAA coverage and three ISO standards. Memory inside the seal is not stated. | Runbook move 4 |
| Q5What leaves with the customer: identity, memory, records, and tuned models? | Not stated No export path is documented for agent state or for licensed weights. | Runbook move 5 |
| Q6What do the 50 million agents count, and what audit record does each leave? | Stated in part Microsoft says registered since May 2026. Autopilot’s audit record is not guaranteed complete. | Runbook move 6 |
Table 5 tracks six series-level questions only Microsoft can answer against what it had stated by 1:20 a.m. UTC on November 19, 2026, and shows two answered in part and four not at all. Answered in part: whether every credit-using service and Perception will have a published rate, and what the 50 million agents count and what audit record each leaves. Not stated: when the agent features reach general availability and on which terms, which controls are enforced on each platform, where agent memory and Frontier data rest, and what leaves with the customer. Each question names the Runbook move it settles. The grading is Greyhound Research analysis.
Two answers would change this note’s reading most. One is a documented rule at the tool call, with a list of what Agent 365 can stop on each platform. The other is general availability terms for the agent features, with their coverage. Either would move part of the control back toward Microsoft’s product. None of the six is a trick question. Each asks Microsoft to write down what its products do or will do at release.
The Deeper Read · About 2 minutes
Select a question to open its full text
Q1When will the agent features reach general availability, and on which terms?
Said so far. In part for a few items. Copilot Code is due in December, and connected platforms complete in December. Autopilot, Workspaces, Today, Agent Grid, the AKS sandbox extension, the confidential computing items, and Licensed Weights have no date. The answer settles. When a pilot can become a production commitment, under the first Runbook move. Evidence that would settle it. Release dates with service levels and compliance coverage. Greyhound Research would update. Exhibit 1 and Table 1.
Q2Will every credit-using service, and Perception, have a published rate?
Said so far. In part. Seat prices are held and published. Several credit rates and a Perception rate card are not. The answer settles. Whether the spending policy of the second Runbook move can be a forecast. Evidence that would settle it. One published rate card with caps by unit. Greyhound Research would update. Exhibit 3.
Q3For each platform, which controls are enforced and which are only observed?
Said so far. Not stated as a list. Microsoft’s documentation calls the link to rivals’ platforms read-only. The answer settles. Where an agent can be stopped, under the third Runbook move. Evidence that would settle it. A per-platform enforcement matrix and a documented rule at the tool call. Greyhound Research would update. Exhibit 4.
Why read-only matters. The phrase is short and easy to pass over, but it already answers half of the question. A read-only link means Agent 365 can report on a rival’s agent and cannot change or halt it. The open half is whether that will change, and on which platforms first.
Q4Where do agent memory, cached context, and Frontier data rest, and under which terms?
Said so far. Not stated. Frontier terms leave out HIPAA coverage and three ISO standards, and Microsoft has not said whether agent memory sits inside the confidential seal. The answer settles. Which data places the fourth Runbook move can mark as written. Evidence that would settle it. A data boundary and retention statement for each agent feature. Greyhound Research would update. Exhibit 4 and Table 3.
Q5What leaves with the customer: identity, memory, records, and tuned models?
Said so far. Not stated. The answer settles. What the exit and model terms of the fifth Runbook move can rely on. Evidence that would settle it. A documented export path for agent identity, state, and audit records, and terms for licensed weights. Greyhound Research would update. Exhibit 3.
Q6What do the 50 million agents count, and what audit record does each leave?
Said so far. In part. Microsoft says the agents were registered since the May 2026 launch, and its Autopilot documentation does not guarantee a complete audit record. The answer settles. What the independent audit of the sixth Runbook move can sample. Evidence that would settle it. A defined count of active agents and a documented audit record for each agent type. Greyhound Research would update. Exhibit 5.
Part 6The Final Greyhound Standpoint
Microsoft Ignite 2026: The Final Greyhound Standpoint
Greyhound SpotlightFour signals within twelve months would move this position, and Microsoft controls all four.
Greyhound Research’s position is that Ignite 2026 made AI agents part of the enterprise workforce, and that this changes where control sits. Microsoft gave agents names, machines, budgets, sealed data, and models of its own, and made Copilot the door every employee walks through. When agents join the workforce, written authority becomes the control: who may enroll, spend, act, and leave, set down by the buyer before the agent starts work. That is not a criticism of the products. It describes what any capable tool needs before it can be trusted with real work.
The judgment rests on a gap between product and terms. On products, Microsoft leads on the breadth of what it ships and on the identity, security, and Microsoft 365 estate around the agent, as the companion notes find area by area. On terms, it trails its own products. The agent features sit in Frontier and preview, several rates are unpublished, and the rule at each tool call is not yet written. The tools are ready to enforce rules that many companies have not yet drafted, and the cost of that gap lands with the buyer, as unplanned spend, unclear accountability, or data held outside written terms.
Earlier platform shifts suggest the gap is normal and, in time, temporary. Client-server computing, software as a service, and the first wave of public cloud each shipped capability ahead of the contracts, rate cards, and controls that later made it routine. In each, the buyers who fared better tended to be those who adopted what was released and wrote their own rules for the rest, rather than those who waited or those who signed on the vision. Microsoft’s record over the past year supports that approach. It turned the previews of Ignite 2025 into priced, released products, which justifies planning on its stated dates while still contracting only on released terms.
The evidence has limits. This note summarizes eight companion notes and relies on their reading of Microsoft’s documentation, which each sets out in full. Most agent features are previews, and Microsoft’s pages word some statuses differently. Company figures, including the 50 million agents and the cost claims for MAI models, are Microsoft’s own. Statuses and claims are as each company published them through 1:20 a.m. UTC on November 19, 2026.
Four signals would move this position within twelve months. Each would move part of the control back into Microsoft’s product. They are general availability terms for the agent features, one published rate card, documented enforcement where the agent runs, and written boundaries for agent data with an export path. Exhibit 6 sets those signals side by side with what each would change, and all four sit with Microsoft.
Greyhound Research · Exhibit 6 · Signals that would move the position
Four signals in twelve months would move this position
What would move Greyhound Research’s series-level judgment within twelve months, and what each would change.
Enforcement where the agent runs is the signal that would move this view furthest. Watch it first.
Exhibit 6 lists four signals that would move Greyhound Research’s series-level position within twelve months, each controlled by Microsoft, and one event outside its control. General availability terms for the agents would let buyers widen enrollment. One rate card for credits and Perception would make the spending policy a forecast. Enforcement where the agent runs, at the tool call and on each platform, would turn the directory into a control. Written boundaries for agent memory and an export path would settle where agent data rests and what leaves with the customer. A rival’s announcement could reset the comparison. The signals are Greyhound Research analysis.
One event outside Microsoft’s control could reset the comparison: a rival’s announcement on agent identity, pricing, or enforcement at its own conference. Of the four signals, enforcement where the agent runs would matter most, because it decides whether a name in Microsoft’s directory carries the power to stop. Until the signals arrive, the balanced course is to use what Microsoft has released on its merits and to write down the rules for everything an agent may do. CIOs and boards should make both decisions at the next annual planning cycle, approving the released products for use and the written documents of the ten Runbook moves, which set an agent’s authority, before any agent moves beyond pilot.
Part 7Appendix, Sources, And Disclosures
Appendix: The Evidence Register, Source By Source
Greyhound Research sets out here the evidence behind every fact, figure, and quotation in this research note, one drop-down for each class of source, so that any sentence can be checked against the words it rests on without leaving the page. Each row gives the source, when it was read, the part of the source the words sit in, what it says, and where in this research note that material is used. Times of reading are in UTC. Every source was read on November 18, 2026, and read again on November 19, 2026, between 01:05 and 01:20 UTC, and the companion notes in their final text after their pre-launch audit the same night. Where a passage was summarized and not quoted, the row says so. A passage quoted in The Deeper Read of a section is not repeated here, and the source’s entry names the section that quotes it.
Microsoft announcements and blogs8 sources, 47 extracts
| Source, read, and used | Where in the source | What it says |
|---|---|---|
| Microsoft (2025a)Read November 18, 2026, and again November 19, 2026, between 01:05 and 01:20 UTCUsed in A Year On. Exhibit 2Quoted in The Deeper Read of A Year On | Agent 365 | [quoted in The Deeper Read] |
| Microsoft Entra | [quoted in The Deeper Read] | |
| Microsoft Foundry | "Fabric IQ, now in preview, extends the unified semantic layer" | |
| Microsoft Foundry | [quoted in The Deeper Read] | |
| Databases | A section is headed "Azure HorizonDB, a new PostgreSQL database, in private preview." | |
| Microsoft (2026a)Read November 18, 2026, and again November 19, 2026, between 01:05 and 01:20 UTCUsed in What Microsoft Announced, A Year On. Table 1. Exhibit 1Quoted in The Deeper Read of What Microsoft Announced | Microsoft Perception | [quoted in The Deeper Read] |
| Dynamics 365 | [quoted in The Deeper Read] | |
| Microsoft IQ | "Microsoft IQ, now generally available," The section adds that it comes with no new SKU or subscription (summarized, not quoted). | |
| Databases | "Azure HorizonDB generally available for mission-critical applications" (heading). | |
| Windows | The Windows Endpoint Security Platform, now generally available, gives approved security products a safer way to protect devices (summarized, not quoted). | |
| Reading note | Greyhound Research counted 59 announcement headings, read on November 19, 2026: 10 cover only generally available items, 19 only previews or Frontier items, 28 a mix, and 2 give no status for some or all of their items. Subheadings inside one announcement are not counted separately. | |
| Azure | [quoted in The Deeper Read] | |
| Agent 365 | Cost Management spend visibility, policies, limits, and alerts are in preview, with forecasting in the Frontier program this month (summarized, not quoted). | |
| Reading note | The Book of News gives Cobalt 200 a footprint of 24 regions by year end, while Microsoft Source gives 23. This note uses neither count. | |
| Microsoft (2026b)Read November 18, 2026, and again November 19, 2026, between 01:05 and 01:20 UTCUsed in What Ignite Changed, What Microsoft Announced, The Through-Line, The Defender, The Model, And The ExitQuoted in The Deeper Read of What Ignite Changed, The Through-Line, The Defender, The Model, And The Exit | Byline | The post is credited to Microsoft Corporate Blogs and presented as an edited transcript of Satya Nadella's opening keynote (summarized, not quoted). |
| Keynote | [quoted in The Deeper Read] | |
| Keynote | [quoted in The Deeper Read] | |
| Keynote | [quoted in The Deeper Read] | |
| Keynote | [quoted in The Deeper Read] | |
| Keynote | [quoted in The Deeper Read] | |
| Keynote | [quoted in The Deeper Read] | |
| Keynote | [quoted in The Deeper Read] | |
| Keynote | Maia 300 is described as delivering 40 percent more tokens per watt than leading GPU-based systems, a company claim not used in the analysis (summarized, not quoted). | |
| Keynote | "We already have over 50 million agents in Agent 365." | |
| Microsoft (2026c)Read November 18, 2026, and again November 19, 2026, between 01:05 and 01:20 UTCUsed in What Ignite Changed, What Microsoft Announced, A Year On, The Front Door And The Bill, The Defender, The Model, And The ExitQuoted in The Deeper Read of A Year On, What Ignite Changed, The Front Door And The Bill, The Defender, The Model, And The Exit, What Microsoft Announced | Opening | [quoted in The Deeper Read] |
| Trust | [quoted in The Deeper Read] | |
| Themes | Four themes: human ambition, intelligence, trust, and infrastructure (summarized, not quoted). | |
| Agent 365 | [quoted in The Deeper Read] | |
| Security | [quoted in The Deeper Read] | |
| Infrastructure | [quoted in The Deeper Read] | |
| Microsoft (2026d)Read November 18, 2026, and again November 19, 2026, between 01:05 and 01:20 UTCUsed in A Year On, The Name, The Machine, And The Seal, Questions For MicrosoftQuoted in The Deeper Read of A Year On, The Name, The Machine, And The Seal | Introduction | [quoted in The Deeper Read] |
| Connected platforms | [quoted in The Deeper Read] | |
| Connected platforms | Support is rolling out across Salesforce Agentforce, Amazon AgentCore, Google Vertex AI, Databricks, Snowflake, Anthropic Claude Managed Agents, OpenAI, Oracle, SAP, ServiceNow, and UiPath, completing in December (summarized, not quoted). | |
| Licensing | Premium capabilities, including insights on non-Microsoft agents, still need an Agent 365 or ME7 license (summarized, not quoted). | |
| Althoff, J. (2026)Read November 18, 2026, and again November 19, 2026, between 01:05 and 01:20 UTCUsed in A Year On, The Strongest Objection. Exhibits 2 and 5Quoted in The Deeper Read of A Year On, The Strongest Objection | Post | [quoted in The Deeper Read] |
| Post | [quoted in The Deeper Read] | |
| Post | [quoted in The Deeper Read] | |
| Post | E7 includes Microsoft Entra Suite and advanced Defender, Intune, and Purview security capabilities (summarized, not quoted). | |
| Bhatia, V. (2026)Read November 18, 2026, and again November 19, 2026, between 01:05 and 01:20 UTCUsed in What Microsoft Announced, The Name, The Machine, And The Seal. Exhibits 1 and 4Quoted in The Deeper Read of What Microsoft Announced, The Name, The Machine, And The Seal | Highlights | The Ignite item is described as a confidential bare-metal technology showcase on NVIDIA Vera Rubin NVL72, with a single-node, four-GPU AI demonstration (summarized, not quoted). |
| Status | [quoted in The Deeper Read] | |
| Status | Confidential enclaves and next-generation AKS confidential containers are described as under active development and planned for a future release (summarized, not quoted). | |
| Status | [quoted in The Deeper Read] | |
| Figure 1 | [quoted in The Deeper Read] | |
| Reading note | The Book of News lists three offerings as now in preview. This note follows the detailed post, which is narrower. | |
| Shaw, F. X. (2025)Read November 18, 2026, and again November 19, 2026, between 01:05 and 01:20 UTCUsed in A Year OnQuoted in The Deeper Read of A Year On | Post | [quoted in The Deeper Read] |
| Post | "We envision a future where organizations become Frontier Firms by using AI for unlocking creativity and innovation," | |
| Reading note | A projection of agent numbers in the post is attributed to a third-party study and is not used. |
Partner announcements1 sources, 2 extracts
| Source, read, and used | Where in the source | What it says |
|---|---|---|
| Zenity (2026)Read November 18, 2026, and again November 19, 2026, between 01:05 and 01:20 UTCUsed in The Name, The Machine, And The SealQuoted in The Deeper Read of The Name, The Machine, And The Seal | Release | Ray Smith, Corporate Vice President, Agent 365 Product, Microsoft: [quoted in The Deeper Read] |
| Release | The same quotation continues that Agent 365 helps organizations turn shared visibility into management, governance, and security decisions, and that the Zenity integration brings runtime security evidence through Microsoft Sentinel (summarized, not quoted). |
Media1 sources, 5 extracts
| Source, read, and used | Where in the source | What it says |
|---|---|---|
| Claburn, T. (2026)Read November 18, 2026, and again November 19, 2026, between 01:05 and 01:20 UTCUsed in What Ignite ChangedQuoted in The Deeper Read of What Ignite Changed | Article | [quoted in The Deeper Read] |
| Article | [quoted in The Deeper Read] | |
| Article | Nadella said apps are "just plugins to Copilot" (fragment). | |
| Article | The reporter writes that trust "remains missing in action" (fragment). | |
| Article | The keynote was at Moscone Center in San Francisco (summarized, not quoted). |
The Microsoft Ignite 2026 series8 sources, 16 extracts
| Source, read, and used | Where in the source | What it says |
|---|---|---|
| Greyhound Research (2026a)Final text read November 19, 2026, after its pre-launch audit, before publicationUsed in The Series, The Front Door And The Bill, The Strongest Objection, The Greyhound CXO Runbook | Position | When the front door moves, Frontier enrollment becomes the control. Frontier terms leave out HIPAA coverage and three ISO standards, and Autopilot activity is not a guarantee of a complete audit record (summarized, not quoted). |
| Greyhound Research (2026b)Final text read November 19, 2026, after its pre-launch audit, before publicationUsed in The Through-Line, The Series, The Front Door And The Bill, The Strongest Objection, The Greyhound CXO Runbook | Position | When work is billed by the task, the spending policy becomes the control. A spending limit is a ceiling, not a budget, and several credit rates are unpublished (summarized, not quoted). |
| Partner channel | Usage-based billing becomes the default for new Copilot Business subscriptions bought through partners between January and March 2027, moved from December 1, 2026, with a 4,000-credit default limit per user a month (summarized, not quoted). | |
| Greyhound Research (2026c)Final text read November 19, 2026, after its pre-launch audit, before publicationUsed in A Year On, The Series, The Name, The Machine, And The Seal, The Strongest Objection, The Greyhound CXO Runbook | Position | When rivals' agents get a Microsoft name, enforcement becomes the control. The link to rivals' platforms is read-only, and the 50 million has no published definition (summarized, not quoted). |
| Rivals | AWS Agent Registry has been generally available since August 31, 2026 (summarized, not quoted). | |
| Greyhound Research (2026d)Final text read November 19, 2026, after its pre-launch audit, before publicationUsed in The Through-Line, The Series, The Name, The Machine, And The Seal, The Greyhound CXO Runbook | Position | When agents get their own machines, authority becomes the control. Running agents is level with AWS and Google, AWS leads on tool-level rules, and Microsoft leads on the estate (summarized, not quoted). |
| Context Cache | Processed text is kept for 1 to 30 days counted from last use (summarized, not quoted). | |
| Greyhound Research (2026e)Final text read November 19, 2026, after its pre-launch audit, before publicationUsed in What Microsoft Announced, The Series, The Name, The Machine, And The Seal, The Greyhound CXO Runbook | Position | When agents work inside the seal, attestation becomes the control. The seal keeps the operator out but not the customer's own agent (summarized, not quoted). |
| Status | Of the three Ignite items, one is a limited preview and two remain in development (summarized, not quoted). | |
| Greyhound Research (2026f)Final text read November 19, 2026, after its pre-launch audit, before publicationUsed in The Series, The Defender, The Model, And The Exit, The Greyhound CXO Runbook | Position | When agents close the alerts, the audit becomes the control. Perception's price has a meter but no published rate card (summarized, not quoted). |
| Dates | Project Perception was introduced on July 27, 2026 (summarized, not quoted). | |
| Status | Three Microsoft pages still describe Perception as a preview. CrowdStrike lists approval gates before sensitive tool calls as available today (summarized, not quoted). | |
| Greyhound Research (2026g)Final text read November 19, 2026, after its pre-launch audit, before publicationUsed in The Series, The Defender, The Model, And The Exit, The Greyhound CXO Runbook | Position | When the platform owner ships its own models, model choice becomes the control. MAI grows first inside Microsoft’s own products. It is a credible option for everyday work, likely to become a firm price anchor once list prices are published, and its terms are still catching up (summarized, not quoted). |
| Licensed Weights | MAI-Base-2, described by Microsoft AI on October 15, 2026, is the precursor of weight licensing. Microsoft's blind preference figure now reads 64% on both of its pages (summarized, not quoted). | |
| Greyhound Research (2026h)Final text read November 19, 2026, after its pre-launch audit, before publicationUsed in The Series, The Defender, The Model, And The Exit, The Greyhound CXO Runbook | Position | When leaving gets cheaper, the exit plan becomes the control. Activate has no published price, and its SAP tool is in private preview (summarized, not quoted). |
| Customers | Microsoft names Comcast and Fusion5, a Microsoft partner, among Activate customers. Accenture is design and implementation partner and customer zero for Copilotforce (summarized, not quoted). |
Research this note builds on5 sources, 8 extracts
| Source, read, and used | Where in the source | What it says |
|---|---|---|
| Greyhound Research (2025a)Read November 18, 2026, and again November 19, 2026, at 01:10 UTC, on greyhoundresearch.comUsed in A Year On | Note | Entra Agent ID is described as a policy control plane for agent-based computing (summarized, not quoted). |
| Greyhound Research (2026i)Read November 18, 2026, and again November 19, 2026, at 01:10 UTC, on greyhoundresearch.comUsed in The Through-LineQuoted in The Deeper Read of The Through-Line | Note | [quoted in The Deeper Read] |
| Note | The finding is from Greyhound CIO Pulse 2026, across more than 1,000 enterprise technology leaders (summarized, not quoted). | |
| Greyhound Research (2026j)Read November 18, 2026, and again November 19, 2026, at 01:10 UTC, on greyhoundresearch.comUsed in What Ignite ChangedQuoted in The Deeper Read of What Ignite Changed | Note | [quoted in The Deeper Read] |
| Note | Consolidating on one suite remains right for most enterprises, while the standard terms around it no longer suffice (summarized, not quoted). | |
| Note | By the note's count of the public record, the November 2026 incident is the fourth multi-service Microsoft 365 incident of 2026 (summarized, not quoted). | |
| Greyhound Research (2025b)Read November 19, 2026, at 01:12 UTC, on greyhoundresearch.comUsed in What Ignite Changed | Standpoint | Mitigation is to include backup identity systems and failover messaging tools that integrate with Azure AD and Exchange Online in parallel (summarized, not quoted). |
| Greyhound Research (2025c)Read November 19, 2026, at 02:30 UTC, on greyhoundresearch.comUsed in What Ignite Changed | Standpoint | "At that point, renegotiation leverage is materially diminished." |
Part 7Appendix, Sources, And Disclosures
Sources And References: Every Document Behind Every Claim
Every fact, figure, and quotation in this research note traces to one of the sources below, presented in Harvard reference format and grouped by class. Each source is also linked in the sentence that carries it. Microsoft’s documents are its account of its own products, prices, and terms, and prove what Microsoft said rather than that the thing is true. The media report what their authors saw and were told. The companion notes set out their own sources in full. All sources were read on November 18, 2026, and read again on November 19, 2026, between 1:05 and 1:20 a.m. UTC.
Microsoft announcements and blogs
Microsoft (2025a) Microsoft Ignite 2025 | Book of News. Microsoft News, November 18. Available at: https://news.microsoft.com/ignite-2025-book-of-news/
Microsoft (2026a) Microsoft Ignite 2026 | Book of News. Microsoft News, November 17. Available at: https://news.microsoft.com/events/ignite/2026/book-of-news/
Microsoft (2026b) Ignite 2026: Building the frontier ecosystem. The Official Microsoft Blog, November 17. Available at: https://blogs.microsoft.com/blog/2026/11/17/ignite-2026-building-the-frontier-ecosystem/
Microsoft (2026c) Microsoft Ignite 2026: Empowering human ambition, intelligence and trust. Microsoft Source, November 17. Available at: https://news.microsoft.com/source/2026/11/17/microsoft-ignite-2026-empowering-human-ambition-intelligence-and-trust/
Microsoft (2026d) What’s new in Microsoft Agent 365: Ignite 2026. Microsoft Agent 365 Blog, Microsoft Community Hub, November 17. Available at: https://techcommunity.microsoft.com/blog/agent-365-blog/whats-new-in-microsoft-agent-365-ignite-2026/4571753
Althoff, J. (2026) Introducing the First Frontier Suite built on Intelligence + Trust. The Official Microsoft Blog, March 9. Available at: https://blogs.microsoft.com/blog/2026/03/09/introducing-the-first-frontier-suite-built-on-intelligence-trust/
Bhatia, V. (2026) What’s New in Azure Confidential Computing, Ignite 2026. Microsoft Azure Confidential Computing Blog, Microsoft Community Hub, November 17. Available at: https://techcommunity.microsoft.com/blog/azureconfidentialcomputingblog/what%E2%80%99s-new-in-azure-confidential-computing-%E2%80%94-ignite-2026/4573909
Shaw, F. X. (2025) From idea to deployment: The complete lifecycle of AI on display at Ignite 2025. The Official Microsoft Blog, November 18. Available at: https://blogs.microsoft.com/blog/2025/11/18/from-idea-to-deployment-the-complete-lifecycle-of-ai-on-display-at-ignite-2025/
Partner announcements
Zenity (2026) Zenity expands AI agent security coverage across the Microsoft ecosystem. Business Wire, November 17. Available at: https://www.financialcontent.com/article/bizwire-2026-11-17-zenity-expands-ai-agent-security-coverage-across-the-microsoft-ecosystem
Media
Claburn, T. (2026) Satya sorry for SaaS-y SNAFU - look over there, more AI! The Register, November 17. Available at: https://www.theregister.com/ai-and-ml/2026/11/17/satya-sorry-for-saas-y-snafu-look-over-there-more-ai/5310594
The Microsoft Ignite 2026 series
Greyhound Research (2026a) Microsoft Copilot As The OS For Work – When The Front Door Moves, Frontier Enrollment Becomes The Control. Research note, November 2026. Available at: https://greyhoundresearch.com/microsoft-copilot-as-the-os-for-work-when-the-front-door-moves-frontier-enrollment-becomes-the-control/
Greyhound Research (2026b) Microsoft Agent Economics – When Work Is Billed By The Task, The Spending Policy Becomes The Control. Research note, November 2026. Available at: https://greyhoundresearch.com/microsoft-agent-economics-when-work-is-billed-by-the-task-the-spending-policy-becomes-the-control/
Greyhound Research (2026c) Microsoft Agent Identity – When Rivals’ Agents Get A Microsoft Name, Enforcement Becomes The Control. Research note, November 2026. Available at: https://greyhoundresearch.com/microsoft-agent-identity-when-rivals-agents-get-a-microsoft-name-enforcement-becomes-the-control/
Greyhound Research (2026d) Microsoft Azure Agent Infrastructure – When Agents Get Their Own Machines, Authority Becomes The Control. Research note, November 2026. Available at: https://greyhoundresearch.com/microsoft-azure-agent-infrastructure-when-agents-get-their-own-machines-authority-becomes-the-control/
Greyhound Research (2026e) Microsoft Confidential Computing – When Agents Work Inside The Seal, Attestation Becomes The Control. Research note, November 2026. Available at: https://greyhoundresearch.com/microsoft-confidential-computing-when-agents-work-inside-the-seal-attestation-becomes-the-control/
Greyhound Research (2026f) Microsoft Perception – When Agents Close The Alerts, The Audit Becomes The Control. Research note, November 2026. Available at: https://greyhoundresearch.com/microsoft-perception-when-agents-close-the-alerts-the-audit-becomes-the-control/
Greyhound Research (2026g) Microsoft MAI Models – When The Platform Owner Ships Its Own Models, Model Choice Becomes The Control. Research note, November 2026. Available at: https://greyhoundresearch.com/microsoft-mai-models-when-the-platform-owner-ships-its-own-models-model-choice-becomes-the-control/
Greyhound Research (2026h) Microsoft Dynamics 365 Activate – When Leaving Gets Cheaper, The Exit Plan Becomes The Control. Research note, November 2026. Available at: https://greyhoundresearch.com/microsoft-dynamics-365-activate-when-leaving-gets-cheaper-the-exit-plan-becomes-the-control/
Research this note builds on
Greyhound Research (2025a) Microsoft Build 2025: Making AI Agents More Capable and Secure. Research note, May 22. Available at: https://greyhoundresearch.com/microsoft-build-2025-making-ai-agents-more-capable-and-secure/
Greyhound Research (2025b) Microsoft 365 Outage on June 17, 2025: What Happened? Research note, June 18. Available at: https://greyhoundresearch.com/microsoft-365-outage-on-june-17-2025-what-happened/
Greyhound Research (2025c) How Microsoft’s Researcher And Analyst Copilot Agents Are Transforming Enterprise Workflows With AI Reasoning. Research note, July 14. Available at: https://greyhoundresearch.com/microsofts-copilot-gets-a-brain-what-researchers-and-analysts-mean-for-enterprise-workflows/
Greyhound Research (2026i) The Front Door Race – Oracle, OpenAI And Five Rivals, And Why The Permission Becomes The Control. Research note, November 5. Available at: https://greyhoundresearch.com/oracle-openai-the-front-door-race-the-permission-becomes-the-control/
Greyhound Research (2026j) Microsoft 365 Outage – When The Suite Goes Dark, Restoration Order Becomes The Control. Research note, November 15. Available at: https://greyhoundresearch.com/microsoft-365-outage-when-the-suite-goes-dark-restoration-order-becomes-the-control/
Part 7Appendix, Sources, And Disclosures
Disclosures: Independence, Financial, Intended Use, Methodology, Authoring, And Copyright
The evidence in this research note is current through November 19, 2026, with an evidence cutoff of 1:20 a.m. UTC that day (8:20 p.m. ET on November 18), after a pre-launch re-verification of every source. The companion notes it summarizes each state their own cutoff. Any written response received from an organization named in it after that time is dated where it appears.
Independence Disclosure
Research under the Greyhound Research brand is independent, and no organization holds a right of approval over its content or publication. It is not commissioned, sponsored, or funded by any organization it covers. Greyhound Research may send a research note to companies named in it for factual review before publication. Any such review is confined to the correction of fact, and no viewpoint, conclusion, or recommendation is guided by those companies. All insights and recommendations are Greyhound Research’s alone, and editorial control remains entirely with Greyhound Research. Greyhound Consilium is the B2B thought-leadership arm of The House Of Greyhound and a sister brand to Greyhound Research. It builds on Greyhound Research’s independent research through collaborations that produce thought-leadership content, research, and practical toolkits for decision-makers. Every Greyhound Consilium publication names its partner on the face of the document. This research note is published under the Greyhound Research brand.
Financial Disclosure
Greyhound Research undertakes advisory and consulting engagements with enterprise buyers and technology providers, and no such engagement confers control over its research findings or its decision to publish them. For this research note, Greyhound Research has no financial interest in, and no commercial relationship with, Microsoft, Amazon Web Services, Google, Anthropic, or OpenAI to declare. Oracle is a client of Greyhound Research. Oracle is not a subject of this research note.
Intended Use
This research note is written for chief information officers, chief financial officers, chief information security officers, data protection officers, chief risk officers, procurement and vendor management leaders, general counsel, board members, and the press. It summarizes Greyhound Research’s eight-note series on Microsoft Ignite 2026 and reads Microsoft’s Ignite announcements, keynote, and product blogs, with Ignite 2025 as the baseline, to inform decisions on agents, Copilot, spending, identity, security operations, models, and contracts. This research note is not investment advice, a recommendation to buy, sell, or hold any security, or a view on any company’s share price or valuation. It is not legal advice, and its reading of contract and data protection matters is to be confirmed with counsel. Its assessments rest on the documents listed in Sources And References as they stood at 1:20 a.m. UTC on November 19, 2026, and can change as the companies named publish more. Product availability, status, features, and prices are described as each company published them through 1:20 a.m. UTC on November 19, 2026, so confirm them with the vendor before acting on them.
Research Methodology
Greyhound Research’s analysis combines lived experience with public sources. The lived experience is years of continuously tracking the market, Greyhound Fieldnotes from ongoing advisory engagements with enterprise buyers and with technology providers, and Greyhound Pulse, the firm’s market trackers. Lived experience forms the judgment and never stands in for evidence. The public sources are the twenty-three documents listed in Sources And References, in five classes, and every factual claim is verified against the document that supports it. The positions of the eight companion notes are summarized from their final texts, and their evidence is set out in each note. The passages relied on are quoted in The Deeper Read of the section that uses them, and the other words taken from each source are set out in the appendix. A finding that something is not stated or not documented is bounded to the documents the companion notes name, as they stood when each was last read on November 19, 2026. Statuses are given as each company states them.
The ranking of announcements in Table 1 is Greyhound Research’s judgment of buyer consequence. Two dates in Exhibit 2, Cowork’s general availability on June 16, 2026, and the introduction of Project Perception on July 27, 2026, are carried from the companion notes that verified them. Company figures, including the 50 million registered agents and Microsoft’s performance and cost claims, are Microsoft’s own and are not used as measures. The Greyhound CIO Pulse 2026 figure is cited as published in Greyhound Research’s research note of November 5, 2026. The count of Book of News headings in What Microsoft Announced is Greyhound Research’s own, made from the Book of News as read on November 19, 2026. No composite figure is computed. No Greyhound Fieldnotes are quoted, and no factual claim in this research note rests on unpublished material.
The through-line, the readings in Exhibits 1 to 6 and Tables 1 to 5, the risks, the Greyhound CXO Runbook, the questions for Microsoft, and the signals are Greyhound Research’s analysis, not statements by Microsoft or any other company. Quotations are checked against the live source before publication.
Authoring Disclosure
This material reflects the analysis, viewpoint and editorial judgment of Greyhound Research. The position presented is developed from Greyhound Research’s research, advisory work and accumulated institutional knowledge. Depending on the assignment, AI may be used to retrieve relevant material; structure, draft or polish prose from analyst-provided concepts, dictation, notes, rough drafts and source material; support secondary research and factual cross-checking; and assist with editing, condensation and consistency with Greyhound Research’s house style. AI does not independently establish Greyhound Research’s viewpoint, analytical position or conclusions. Material factual claims are verified against source evidence before release. Greyhound Research reviews and approves the final material and retains responsibility for it.
Copyright Policy
All content in this document and on the Greyhound Research website is protected by copyright law and may not be reproduced, distributed, transmitted, displayed, published, or broadcast without the prior written permission of Greyhound Research or, in the case of third-party materials, the prior written consent of the copyright owner of that content. You may not alter, delete, obscure, or conceal any trademark, copyright, or other notice appearing in any Greyhound Research content. Quotation of up to one hundred words is permitted where it carries attribution to Greyhound Research and a link to the published page. Reproduction, redistribution or republication beyond that extent, in whole or in part, by email or in any medium including websites, newsletters and intranets, requires prior written permission, which may be requested at the address below. The published version of each content piece on the Greyhound Research website carries sharing tools for readers who wish to share it.
For questions arising from these disclosures, including licensing requests, contact the Greyhound Research Community Relations Team at connect@thofgr.com. For media inquiries, write to media@thofgr.com.

Analyst In Focus: Sanchit Vir Gogia
Sanchit Vir Gogia, or SVG as he is popularly known, is a globally recognized technology analyst, innovation strategist, digital consultant and board advisor. SVG is the Chief Analyst at Greyhound Research (part of The House Of Greyhound), a Global, Award-Winning Technology Research, Advisory, Consulting & Education firm. Greyhound Research works closely with global organizations, their CxOs and the Board of Directors on Technology & Digital Transformation decisions.
Discover more from Greyhound Research
Subscribe to get the latest posts sent to your email.
