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 Oracle AI World, Oracle brought its rivals inside its own cloud. Google’s TPUs, Cerebras inference, SambaNova, Intel and AMD chips, and OpenAI’s own Jalapeño accelerators are coming to OCI, beside NVIDIA’s and Positron’s systems, which Oracle says are available now. Snowflake will run as a first-party service, bought and billed through Oracle, with Oracle Universal Credits usable. In the other direction, Oracle’s database is now in seventeen Google Cloud regions, including Mexico, and Oracle plans FedRAMP Class D certification for it in the government clouds of AWS, Microsoft and Google. Of fourteen chips, platforms and cloud services announced, five are available now, two are dated for 2027, and seven have no date.
Greyhound Research reads the announcements as the point at which Oracle completed a turn it began years ago: many vendors’ technology on one Oracle contract, and Oracle’s database inside every major cloud. The architecture no longer sets the boundaries between vendors. The contract does. The releases write the commercial path clearly, how to buy and who bills, and leave five terms to the buyer: who supports a rival’s service, where it runs, what the money buys and where it stops, when it arrives, and how a workload leaves. Oracle’s documents name no first responder for any rival service, and no region for any of the new chips or for Snowflake.
OpenAI shows the model at its clearest, as a tenant of OCI, a maker of chips planned for it, and a supplier to Oracle’s customers on Oracle’s credits, and where OpenAI models bought on those credits run is not stated. Oracle’s own page draws the edges of its Multicloud Universal Credits: USD only, not available in Government or Alloy data centers, and list price for Database@ use if the contract is not renewed. The test for buyers is whether every rival service bought through Oracle comes with a named first responder, a written region, a credit plan that survives the year, a date they can plan on, and a way out they have tried. The research note sets out a six-move Greyhound CXO Runbook, one move for each leader, with twelve exhibits, five tables and eight questions for Oracle.
In This Research NoteSection guide · AI drafted · Analyst reviewed
Nineteen sections, each with a short summary. Select a title to go straight to that section.
Oracle used AI World to bring its rivals inside its own cloud and take its database further into theirs. Greyhound Research reads it as the point at which the contract becomes the control, and sets the test for every rival service bought through Oracle.
The infrastructure and multicloud announcements, read by status in Oracle’s own words. Of fourteen chips, platforms and cloud services, five are available now, two are dated for 2027, and seven have no date.
Eight makers’ chips are on, or coming to, OCI. NVIDIA’s and Positron’s are available now, Google’s TPUs and Cerebras inference have a year, and no release names a customer region for any chip.
Oracle Acceleron’s four network capabilities carry no date, and the operations tools are documented for Oracle’s GPU nodes. Whether they reach TPUs, Cerebras or SambaNova systems is not stated, and partners appear at the edges.
OpenAI is a tenant of OCI, a maker of chips planned for it, and a supplier on Oracle’s credits. Where a buyer’s OpenAI models bought on those credits run is the one thing not written.
Snowflake is to be bought, provisioned and billed through Oracle, with Universal Credits usable. Who takes the first call, whose service level applies, and the date, regions and price are not stated.
Every offer read against support, place, money, time and exit. For the rival services, the documents write the money and sometimes the date, and leave support, place and exit to the buyer.
Each rival service has two companies behind it and no document saying which stands first. Greyhound Research sets out seven things a support schedule for a rival service should say, from the first responder to what happens if the partnership changes.
One Oracle commitment now reaches Snowflake, OpenAI, Marketplace partners and Oracle’s database in three other clouds. Oracle’s own page sets the limits: USD only, not in Government or Alloy data centers, and list price for Database@ use if not renewed.
Oracle’s database now runs in seventeen Google Cloud regions, and named customers include Banco Actinver, ADT and Vodafone. The data and the service travel, and the AI beside it is the host cloud’s.
No release names a region for the new chips or Snowflake. Oracle’s Gemini notes show the model to follow, answering where data rests, where it is processed, and whose infrastructure processes it.
Oracle plans FedRAMP Class D for its database in three US government clouds, with no date. Which AI services beside it are authorized is not stated, and Oracle’s credits and some commercial models are not yet going with it.
Two chips have a year, and Snowflake, Intel, AMD, SambaNova and NVIDIA’s next eight-GPU system have none. Four conditions for signing for a calendar, and five ways out: one described, one planned, one with a written cost, and two not described.
Six moves for the CIO, the head of infrastructure, vendor management, the CFO, the CISO and the chief data officer, each naming who decides. Four of the six start now.
Eight questions, each led by a buyer signal, on support, regions, dates, where OpenAI models run, the credits, the government stack, portability and the common layers. Each maps to a Runbook move a buyer can make meanwhile.
Oracle has turned multicloud inside out and supplied the choice. Buyers who treat the contract as the control gain most, and ten signals would move the position, four of them in the buyer’s own estate.
Forty-four sources in seven classes, from Oracle’s announcements, blogs, documentation and investor day to partner announcements, named media, and the research this note builds on.
Independence first, including that no named organization was offered a pre-publication factual review. The Financial Disclosure states that Oracle is a client of Greyhound Research and that the engagement did not fund this note. Intended use, research method, authoring, and copyright terms follow.
What AI World Changed: Multicloud Turns Inside Out
Oracle used AI World to bring its rivals inside its own cloud. Google’s Tensor Processing Units are coming to Oracle Cloud Infrastructure (OCI), and Oracle’s release said this is the first time Google TPUs will be offered on a hyperscale cloud platform outside Google. Another release said Snowflake will run on OCI as a native, first-party service, bought and billed through Oracle. And Karan Batta, senior vice president, product, Oracle Cloud Infrastructure, wrote that Oracle plans to deploy OpenAI’s own Jalapeño inference accelerators to add efficient inference capacity for OpenAI.
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 Oracle AI World, Oracle presented Fusion Claw function by function. Its four releases around the event named sixteen Claw-powered applications across finance, HR, supply chain and customer experience, of the twenty-five Oracle says are available. Seven of the sixteen were new names, and six of the seven are in HR, in pay, payroll, hiring and positions. Oracle’s HR release states that organizations retain responsibility for employment decisions. No release gives a date, region or price, and Oracle’s documentation places the applications in Release 26D.
Greyhound Research reads the announcements as the point at which Fusion Claw stops being one decision. Each application sits on a Fusion module with its own approvals, its own way back, its own calendar and, in HR, its own law, so each needs a named owner who signs for what it commits. Oracle’s releases describe every application with the point at which a person comes in. Its 26D documentation, where it exists, is more cautious than the releases in HR and customer experience, and more autonomous in finance. Five of the sixteen have a 26D page that describes them as Claw-powered, and none of the six new HR names has one yet.
The hardest questions sit in the signatures, the handoffs and the calendar. Where an approver runs an application under their own login and then approves its recommendation, the trail can show one identity on both sides. One approved action can cross three to five owners, and no outcome record yet links the steps. The update arrives in the busiest weeks of the year for payroll, finance and supply chain. The research note sets out a six-move Greyhound CXO Runbook, one move for each leader, with twelve exhibits, five tables and seven questions for Oracle.
In This Research NoteSection guide · AI drafted · Analyst reviewed
Sixteen sections, each with a short summary. Select a title to go straight to that section.
Oracle used AI World to present Fusion Claw function by function. Greyhound Research reads it as the point at which Fusion Claw stops being one decision, and sets the test: a named owner for every application.
The AI World releases name sixteen applications, seven of them new names and six of those in HR. HR now has ten names, the most of any function, and finance still has two.
Oracle’s release wording writes four applications in full-auto terms, four to review and then execute, seven to stop at a recommendation, and one to produce a document. Where a 26D page exists, it can promise less than the release, or more.
Ledger is the one finance application, and ten of the sixteen descriptions name money. A Claw recommendation can reach its approver three ways, and in one of them the same identity prepares and approves.
Each HR application lands on a Fusion module with a way back and a point after which the change has reached a person. The EU AI Act, New York City’s law, US state rules and the EU Pay Transparency Directive reach the decision applications first.
Each of the three applications ends in a commitment outside the company, and each way back narrows once the other party has it. Greyhound Research sees supply chain as the function ready first.
The Fusion modules beneath the five applications pass a change on by themselves, so the checkpoint belongs before the change. No customer has yet spoken on the record about running one.
Four chains in which one application’s approved action becomes another function’s input. Each crosses at least three owners, and no outcome record yet links the steps.
Budgets per application, warning thresholds, auto-stop and a stop for all items map onto ownership by function. Table 4 sets out who signs, who co-signs and who holds each budget.
Applying Oracle’s cohort rule puts Release 26D in production on November 20, 2026, December 18, 2026, or January 15, 2027, and payroll environments take the first. It lands well for sales planning and in the busiest weeks for payroll, finance and supply chain.
Six risks, from one identity preparing and approving to cost and chains scattered across four owners. Each remedy sits with a role the buyer already has.
Six moves, one for each leader, from an owner for every application and chain to one authority model, one total budget and one stop. Four of the six start now.
Seven questions, each led by a buyer signal, on names and cohorts, run identity, pay criteria, outreach, ways back, linked records and production customers. Each maps to a Runbook move a buyer can make meanwhile.
Oracle has taken Fusion Claw into every function and described each application with the point at which a person comes in. Buyers who treat ownership as the control gain most, and ten signals would move the position.
Eighty-four sources in seven classes, from Oracle’s announcements, documentation, blogs and investor day to named media, the rules and official texts cited, and the research this note builds on.
Independence first, including that no named organization was offered a pre-publication factual review. The Financial Disclosure states that Oracle is a client of Greyhound Research and that the engagement did not fund this note. Intended use, research method, authoring, and copyright terms follow.
What AI World Changed: Fusion Claw Stops Being One Decision
Oracle used AI World to present Fusion Claw function by function. Its release on Fusion Applications said twenty-five Claw-powered agentic applications are now available across its finance, HR, supply chain and customer experience suites, and the four releases it issued around the event named sixteen of them. Chris Leone, executive vice president, Applications Development, at Oracle, said the next era of enterprise AI is about delivering software that increasingly does the work, with people defining the objectives and setting the policies.
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 Oracle AI World, OpenAI’s chief executive told enterprises that the one thing to do with AI now is to understand the security threat it creates, and Oracle’s chief executive said AI is surfacing flaws built up over 50 years. Oracle announced Oracle AI Security Mesh, powered by OpenAI, a managed service that finds and fixes vulnerabilities with the customer’s approval. Beside it sit OCI Identity Assurance, biometric proof for the people who approve sensitive actions, and OCI Zero Trust Packet Routing, network policy enforced beneath applications and agents. Oracle had already moved its own patching to a monthly cycle, and its largest security release credited AI for the findings.
Greyhound Research reads this as the right diagnosis, and as a shift that moves the bottleneck rather than removing it. When AI finds flaws faster, the scarce resource is the capacity to fix them. Oracle’s layered design places a control at each layer an attack can cross, from the approver to the data, and AI Security Mesh describes the strictest agent identity model in Oracle’s announcements: a distinct identity, task scope, and just-in-time permissions. AI Security Mesh carries no date, region, or price, and where its models run and how findings are held are not yet described.
For buyers, the gains depend on four things they control: a patch cadence that matches discovery, one identity standard for every agent, stronger proof for the people who approve high-risk actions, and controls that sit beneath the agents. The research note sets out ten terms to settle before an attacking agent tests production, six high-risk approvals and the checks each needs, five risks with their early warnings, and a six-move Greyhound CXO Runbook. It carries eight exhibits, three tables, and six questions for Oracle.
In This Research NoteSection guide · AI drafted · Analyst reviewed
Fourteen sections, each with a short summary. Select a title to go straight to that section.
Oracle and OpenAI put AI security first at Oracle AI World. Greyhound Research reads the announcements as a correct diagnosis, and as a shift of the bottleneck from finding flaws to fixing them.
Three offerings sit beside a cadence already in place. Monthly patching and network policy are in effect, and AI Security Mesh carries no date, region, or price.
Oracle moved its own patching to a monthly cycle and shipped its largest security release before announcing a service to find flaws faster. The capacity to fix now sets the pace.
A control at each of five layers, from the approver to the code, with a person’s approval before AI Security Mesh tests or fixes anything. The newest layer is the one a buyer cannot yet date.
AI Security Mesh describes agents with a distinct identity, task scope, and just-in-time permissions. Fusion Claw borrows the user’s identity, and the implementation agents’ identity is not described.
Ten terms to settle before an attacking agent tests production. Oracle describes four, four more in part, and leaves where the models run and what they see undescribed.
Agents move an attacker’s attention to the people who approve. Six high-risk approvals and the checks each deserves, from verified presence to a second signer and a ceiling on volume.
More code is written by AI, and more flaws are found by AI. Oracle describes controls at four of six stages, and leaves review and patching of a buyer’s own generated code to the buyer.
Six moves with named owners, from the monthly patch cycle to an inventory of AI-written code. Four start now, and none waits for AI Security Mesh to carry a date.
Six questions, each led by a buyer signal. Five concern terms Oracle describes in part or not at all, and each maps to a Runbook move a buyer can make meanwhile.
Oracle has read the AI-era security problem correctly and answered it with a sound design, and the evidence is uneven. Buyers who treat fixing as the control gain most, and six signals would move the position.
Twenty-four sources in five classes, from Oracle’s announcements, product page, and blogs to named media and the research this note builds on, each linked where it appears.
Independence first, including that no named organization was offered a pre-publication factual review. The Financial Disclosure states that Oracle is a client of Greyhound Research and that the engagement did not fund this note. Intended use, research method, authoring, and copyright terms follow.
What AI Has Changed: Flaws Found Faster Than They Can Be Fixed
In a discussion with Oracle chief executive Clay Magouyrk at Oracle AI World, Sam Altman, chief executive of OpenAI, said that if he could recommend only one thing for a company to do with AI right now, in the very short term, it would be to “really understand the security threat landscape from AI and mitigate against that.” He added that the sheer number of vulnerabilities AI systems are finding has shocked him.
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 Oracle AI World, Oracle set out to shorten the wait between signing an enterprise application contract and seeing value. It announced Oracle AI Success Navigator, six agents that help with discovery, design, configuration, testing, integration, and adoption in Fusion implementations, and the Oracle Fusion Setup and Configuration Registry, a public, machine-readable source of how Fusion should be set up. It also announced SuiteExpress, an agentic implementation toolset for NetSuite, and App Composer, which lets employees build applications with OpenAI’s Codex. The registry, SuiteExpress and App Composer carry stated availability, SuiteExpress only for software-industry customers in the United States and Canada and App Composer in limited availability.
Greyhound Research reads this as the right target, and as a shift that moves risk rather than removing it. Configuration holds the approvals, roles, and limits that Fusion Claw and other agentic applications later obey, so whoever signs off agent-drafted configuration is signing off authority. Oracle’s releases describe human approval at two of six stages, Design and Configuration, and do not yet describe the identity the agents act under, a record of their work, or rollback of applied configuration. Customers at the event put process, data, and legal readiness first, and Lexington Health’s self-reported 80 percent reduction in certain design and configuration efforts is the strongest customer evidence so far.
For buyers, the gains depend less on the agents than on four things they control: the readiness of their processes, data, and legal review, how the implementation contract is written, who signs off what the agents draft, and who owns what gets generated. The research note sets out a roles grid in which no drafter approves its own output, nine statement-of-work clauses, four risks with their early warnings, and a six-move Greyhound CXO Runbook in which every move starts before the first agent runs. It carries nine exhibits, four tables, and six questions for Oracle.
In This Research NoteSection guide · AI drafted · Analyst reviewed
Fourteen sections, each with a short summary. Select a title to go straight to that section.
Oracle set out to shorten the wait between contract and value, with agents that draft designs and configuration, a public setup registry, and an app builder that uses Codex. Greyhound Research reads it as the right target and a shift that moves risk rather than removing it.
Five components share one idea but sit at different stages. The registry, SuiteExpress and App Composer carry stated availability, SuiteExpress only for one industry in two countries and App Composer in limited availability.
Oracle customers at the event put process, data, and legal readiness first, and named speed as the payoff. Read as configuration, the approvals they want removed are changes to authority.
The registry is the strongest piece and the one a buyer can test now. Lexington Health reports an 80 percent reduction in certain design and configuration efforts, self-reported from a project still under way.
Configuration holds the approvals, roles, and limits that agentic applications later obey. Oracle’s releases describe review at Design and Configuration, but not identity, a record of agent work, or rollback.
A proposed allocation in which agents draft at every stage, a named person approves at every stage, and the owner of the delegation of authority co-signs changes to who may approve what.
App Composer’s readable code suits enterprise use. Data access, testing at each update, and ownership through retirement are left to the buyer, who has time to set the rules first.
Four risks, from best practice flattening what differentiates to agents carrying bad data further. Three of the four early warnings appear before go-live.
Six moves with named owners, from starting with the registry to giving generated applications a lifecycle. Every move starts before the first agent runs.
Six questions, each led by a buyer signal. Four concern controls Oracle’s releases do not yet describe, and each maps to a Runbook move a buyer can make meanwhile.
Oracle has aimed agentic AI at the right problem from a sound base, and the evidence is early. Buyers who treat setup as governance gain most, and five signals would move the position.
Seventeen sources in five classes, from Oracle’s announcements, documentation, and blogs to named media and the research this note builds on, each linked where it appears.
Independence first, including that no named organization was offered a pre-publication factual review. The Financial Disclosure states that Oracle is a client of Greyhound Research and that the engagement did not fund this note. Intended use, research method, authoring, and copyright terms follow.
What Oracle Set Out To Fix: Implementations That Ask Too Much, Too Early
As NetSuite launched SuiteExpress, Evan Goldberg, founder and executive vice president of Oracle NetSuite, said that long ERP implementations “ask too much of customers” before they can begin to see value. At Oracle AI World, Oracle set out to shorten that wait across both of its application families, with agents that draft designs and configuration, a public registry of how Fusion should be set up, and an app builder that lets employees create applications with OpenAI’s Codex.
A practical guide by Greyhound Consilium, in partnership with Oracle
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.
Guide In Two MinutesGuide synopsis · AI drafted · Analyst reviewed
Drafted by AI from the full guide and reviewed by a Greyhound Consilium analyst before publication. The advice and positions are Greyhound Consilium’s own.
Oracle introduced Fusion Claw on September 29, 2026, a governed runtime that lets Fusion’s agentic applications plan and carry out complex work within the authority a business delegates. Greyhound Consilium developed this guide in partnership with Oracle. It builds on the independent research report Greyhound Research published on October 14, 2026, and turns it into next steps for the CIO and the leaders who own each process in finance, HR, supply chain and sales.
The guide treats Fusion Claw as a dial, not a switch. Oracle’s three settings, human in the loop, human in the lead and autonomous execution, change how often a person steps in, never whether controls apply, because, in Oracle’s words, AI “can’t grant itself authority to act.” The authority sits in controls Fusion already holds: agent roles, tested policy models, approval points, the autonomy setting, budgets and exception routing. Greyhound Consilium’s advice is to set the dial outcome by outcome, start every new outcome at human in the loop, and move it up only as the receipts prove it, with an owner and a second signature for every decision.
Four process cards, for Ledger, Workforce Staffing, Shipping Consolidation and Account Territory Growth Plan, show where the dial starts in each function. A 90-day plan takes the first outcome from a name on a page to a higher setting, and the guide sets out what to measure, how to budget AI Units from your own test runs, who owns each decision, how to read receipts, and what regulators in seven jurisdictions ask. It carries six exhibits, thirteen tools and ten of Oracle’s published examples, and closes with four working sessions with your Oracle account team that turn it into a dated plan.
In This GuideSection guide · AI drafted · Analyst reviewed
Fourteen sections, each with a short summary. Select a title to go straight to that section.
Fusion Claw is a dial, not a switch. The business decides, outcome by outcome, how much authority to hand over, and Oracle’s guardrails apply at every setting. Before anyone opens the settings, settle the outcomes by name, the starting setting, the evidence that moves it up, and who decides.
Most of what a first run needs already exists in Fusion. Six preparations, covering authority, release, budget, data and region, assurance, and people, become fourteen items and two gates in a readiness map.
Choose the first process for the proof it gives: a number the CFO already watches and an executive ready to sign. Let Fusion Claw take the plan, and keep the final commitment with a person at first.
The envelope is set in controls Fusion already holds: roles, tested policy models, approval points and the autonomy setting, with budgets and exception routing keeping each run in hand. A short delegation record names who granted the authority, so every receipt can be checked against it.
Fifteen Claw-powered capabilities give every function more than one place to start. Four process cards, for Ledger, Workforce Staffing, Shipping Consolidation and Account Territory Growth Plan, show where the dial starts and the signal to move it.
Thirteen weeks take the first outcome from a name on a page to a higher setting on the dial, in five steps that follow Oracle’s own 90-day pattern for governing AI.
Business measures show the gain, and control measures show it was earned within the authority granted. AI Units per outcome, measured in your own test runs, set the budget.
Following Oracle’s three layers of ownership, every decision gets an owner and a second name. An action register and a receipt review make the program auditable, and the people whose work changes hear what is changing before the first test run.
Regulators in seven jurisdictions converge on a four-part rule, and Fusion Claw already provides much of it. A regulatory runway sets out each date and what to do, starting with the people questions.
Autonomy expands as receipts prove it: at least one new outcome a quarter, faster for similar ones, with every outcome running under Fusion’s own controls. A quarterly review loop decides what moves up, what holds and what comes next.
Four working sessions with your Oracle account team and implementation partner, all within the first month of the 90-day plan, turn this guide into a dated plan for your first outcome.
Thirty-seven sources in eight classes, from Oracle’s announcements, documentation and blogs to regulators’ texts, employee representatives and named buyers. Each entry carries its link, and every source is also linked where the text uses it.
Independence first, including that the guide was shared with Oracle for review before publication. The Financial Disclosure states that Oracle is a client of Greyhound Research. Intended Use sets out who the guide is for and that it is not legal or investment advice. Research method, authoring and copyright terms follow, including how Oracle and its partners may share the guide.
The Delegation Dial: Authority Granted In Advance, Outcome By Outcome
Every enterprise has work too complex for a simple rule and too frequent to do well by hand: closing the books, staffing a ward, consolidating inbound loads, redrawing sales territories. Teams take it to spreadsheets for hours, and the answer is often stale by the time it is ready.
Report In Two MinutesReport synopsis · AI drafted · Analyst reviewed
Drafted by AI from the full research report and reviewed by a Greyhound Research analyst before publication. The analysis and conclusions are Greyhound Research’s own.
On October 8, 2026, officials announced that the Department of Labor would stop processing pending permanent labor certification applications, known as PERM, and accept no new ones involving eight employers. They named Microsoft, Adobe, and six IT services firms. PERM is the first step in many employment-based green card cases. It is not the H-1B visa, and a PERM suspension does not, by itself, change anyone’s permission to work. As of October 20, 2026, Greyhound Research could not locate the written terms in public channels: a notice, the legal authority relied on, the legal entities covered, or a start date. Notices sent directly to employers would not necessarily appear there. Greyhound Research sets the officials’ words beside the regulation that fits them, the Department’s own case files, and the companies’ statements.
The regulation that best fits the announcement, 20 CFR 656.31, allows processing to be suspended while investigations run, and a suspension is not a finding. Its text, on Greyhound Research’s reading, speaks least directly to refusing new filings, though the Department may rely on an authority it has not yet stated publicly. On Greyhound Research’s count of the Department’s files, the six services firms received 116 PERM certifications in the nine months to June 2026, fewer than Adobe’s 182. Microsoft received 1,600 of the eight’s 1,898, or 84 percent. Of the ten figures officials reportedly cited, three are consistent with public counts, one sits close to a related measure, two are not found in the public record, and four are not tested on it. The government may hold records that are not public, such as investigation files and pending cases.
For CIOs, exposure runs through named people in named services, not brands, and nothing in the officials’ statements, as reported, indicates a product risk for software subscriptions. Suppliers, named or not, face a test of whether their PERM pipelines and their recruitment and pay records would stand scrutiny. Greyhound Research credits enforcement on the programs’ own terms, because it shows buyers whose recruitment and pay records stand up. It treats neither the administration’s public counts nor the companies’ statements as proof of whether any named company replaced US workers. For buyers, the decision rule has four steps: map the business dependency, verify the sponsoring legal entity, obtain the supplier’s case counts by stage for each dependent service, and fund a proportionate continuity response. The full research report carries four exhibits, 22 tables, buyer and supplier playbooks, and 13 questions open on the public record as of October 20, 2026.
In This Research ReportSection guide · AI drafted · Analyst reviewed
Fifteen sections, each with a short summary. Select a title to go straight to that section.
On October 8, officials announced that the Department of Labor would neither process pending PERM cases nor accept new ones involving eight employers. Greyhound Research reads the action as a pause while investigations run, not a finding. As of October 20, 2026, Greyhound Research could not locate the written terms in public channels.
A labor certification belongs to a specific employer and job, so Greyhound Research maps exposure by sponsoring entity and immigration stage, not by brand. Which entities each suspension covers, and when its initial 180 days began, depend on written terms not public as of October 20, 2026.
The H-1B program rests on an employer’s attested word on pay and conditions, while PERM tests whether US workers are available for the job. As announced, only PERM has stopped at the eight, so no H-1B total sizes a supplier’s exposure.
Officials’ figures and company statements are graded against the public record, and each grade describes that record, not records the government may hold. Of the ten figures officials reportedly cited, three are consistent with public counts, one sits close to a related measure, two are not found in the public record, and four are not tested on it.
Among decided cases, the Department’s files show that the six services firms, and Amazon, Google, Meta, and Apple, had largely left PERM by fiscal 2026, while Microsoft had not. Named and unnamed employers sit side by side on each measure tested, so the public files cannot explain why these eight were named. That leaves the officials’ stated basis, alleged fraud and, for Adobe and Microsoft, active federal investigations, as the explanation on the record.
A buyer’s exposure is set by the sponsored people inside its accounts, not by the brands on the list. The section maps sourcing models, the contract levers, who carries the cost of a stalled case, and what a buyer may and may not ask.
The eight share one suspension but not one exposure. On decided cases since fiscal 2025, the largest certification counts are Microsoft’s and then Adobe’s, and the evidence of whether each pipeline would stand scrutiny sits in each supplier’s own records.
Officials tied the list to alleged fraud and, for Adobe and Microsoft, active federal investigations, and Greyhound Research reads the list as silent on every sponsor it leaves out. Settlements, closed wage cases, and audits show that the same recruitment, layoff, and wage questions can reach any sponsor, named or not.
The deadlines inside each worker’s case keep running while processing stands still. Four clocks sit inside the green card case and four in the worker’s status, and three of them can end a case or a stay outright.
A buyer playbook and a supplier playbook, each of eight steps with an owner and evidence that the step was done. Review points from 30 days to 36 months open on evidence, not on elapsed time.
Thirteen questions remain open on the public record as of October 20, 2026, for the Department of Labor, the White House, the named employers, and their peers. Two matter most: the authority cited for refusing new filings and the legal entities each suspension covers.
The suspensions are a delivery problem for buyers, located in named people, and a pipeline test for suppliers. Greyhound Research credits enforcement on the programs’ own terms and sets out what would move its judgment either way.
The record behind the analysis: the supplementary H-1B counts, every legal and immigration point with its official text and practitioner verification, the method and its limits, and the terms used.
Five classes of public sources, 183 in all, begin with official records, company statements and filings, and academic research. Reporting, practitioner interpretation, and organizations’ statements form the fourth class, and prior Greyhound Research analysis the fifth. Each entry carries its link, and every source named in the text is linked where it appears.
Independence first, including that no named organization was offered a pre-publication factual review. The Financial Disclosure states that no organization commissioned, sponsored, or funded the report, including Infosys and Oracle, both clients of Greyhound Research. Intended Use sets out who the report is for and that it is not legal or investment advice. Research method, authoring, and copyright terms follow.
What Changed: The October 8 Suspensions As Announced
At a White House press conference on visa fraud on October 8, 2026, Vice President JD Vance said the government was suspending Microsoft from the program through which employers sponsor foreign workers for permanent residence. He cited alleged fraud. Secretary of Labor Keith Sonderling said the suspension would also cover Adobe and six IT services firms: Cognizant, Infosys, Tata Consultancy Services, Wipro, HCL Technologies (HCLTech), and Capgemini. “We will not accept any new or process any pending permanent labor certification applications involving these companies,” Sonderling said.
Report In Two MinutesReport synopsis · AI drafted · Analyst reviewed
Drafted by AI from the full research report and reviewed by a Greyhound Research analyst before publication. The analysis and conclusions are Greyhound Research’s own.
In July 2026 Microsoft reported that across its global operations in FY25 it replenished more water than it withdrew, 14.28 million cubic meters against 13.27 million. Greyhound Research tests that milestone, and the eleven community commitments Microsoft has made on water, electricity, and local economic return, against one standard applied twice. Does a facility take more than its place can spare, and does the benefit promised in return reach the people asked to accept it? Microsoft’s own foreword accepts the premise, saying that replenishing global volumes is not enough.
Microsoft’s own location table shows where the milestone landed. It lists 29 owned data center sites, about a third of the water the company reports withdrawing. Seven show a replenishment volume, eleven carry a promise dated FY30, and eleven have no plan disclosed. At Quincy, Washington, the largest draw in the table, 1,292 megaliters withdrawn sits against 833 replenished. The milestone sits outside the accountant’s review, and the reviewed withdrawal figure rose 14.5 percent in the same year. On electricity, the renewable record is real and well built, but annual matching cannot show that a local grid had power in the hour a site needed it, and no ratepayer can yet test the price pledge. About 8,550 megaliters of withdrawal sits outside the site table, in leased data centers, offices, and smaller sites, and the economic commitments arrive without baselines. Greyhound Research credits the dated promises, the end of non-disclosure agreements with local governments, and the four sites where replenishment exceeds local withdrawal.
For technology leaders, the consequence runs through specific dependencies, not headlines. Establish which regions and availability zones run on capacity outside Microsoft’s site table before the next renewal, tie capacity plans to utility and approval milestones, and test whether a fallback shares the same grid or water supply. The research report sets out the evidence in ten exhibits and nine tables, the Greyhound CXO Runbook, and the ten questions only Microsoft can answer, most of them for a column or a figure it already holds.
In This Research ReportSection guide · AI drafted · Analyst reviewed
Thirteen sections, each with a short summary. Select a title to go straight to that section.
Every commitment to a community is judged twice: does a facility take more than its place can spare, and does the promised benefit reach the people asked to accept it? Greyhound Research applies the same test to itself.
Eleven commitments from five Microsoft sources, on water, electricity, and local economic return, are now precise enough to test. Exhibit 3 places all eleven on one page, and the pattern repeats: reporting is company-wide, while the promises were local.
Microsoft’s location table covers 29 owned sites and about a third of its reported withdrawals. Seven show replenishment, eleven carry FY30 promises, and eleven show no plan. The global milestone sits outside the accountant’s review.
A local water account needs volumes with local context, benefits dated to the season they arrive, and cooling choices weighed against water and electricity together. The cheapest step, a cooling-type column, is already in Microsoft’s hands.
Microsoft took the harder route to the 100 percent renewable matching it reports for FY25, with about 93 percent from contracts and on-site generation. Annual matching still cannot show hourly local supply, and no ratepayer can yet test the price pledge.
A consultation lasts months, and a data center is expected to run for twenty or thirty years. The record of what was promised needs a stable facility identifier, reporting on an agreed schedule, and a named response when a commitment is missed.
About 8,550 megaliters of Microsoft’s reported withdrawal sits outside the location table, in leased data centers, offices, and smaller sites, with the split unpublished. Greyhound Research holds that a commitment travels with the capacity, whoever operates the building.
Microsoft names amounts, places, and headcounts, and gave up a local incentive at Mount Pleasant, Wisconsin. Without baselines, no one can yet tell what a community gained beyond what it would have received anyway, or which jobs outlast construction.
Five actions for technology, procurement, and risk teams, each with an owner and a moment in the buying cycle. A disclosure gap is not an outage, and the right response depends on which of four kinds of evidence is in hand.
Ten questions only Microsoft can answer, in the order the evidence raises them. Most ask for a column, a scope note, or a figure the company already holds, and two would change what this report can conclude.
The global promise is kept, and the local one is still owed. Greyhound Research credits the dated promises and the openness commitment, and asks for two disclosures: the leased estate in the location table, and delivered volumes at the promised sites.
Forty-two sources in seven classes, from Microsoft’s own disclosures to regulators, scientific references, and independent reporting. Each entry gives the exact title, the date, and a live link, and every source named in the text is linked where it appears.
Independence first, including that this edition was not submitted to Microsoft for factual review. The Financial Disclosure confirms that no organization commissioned or paid for the report and that neither Microsoft nor Amazon is a client, Intended Use sets out who the report is for and that it is not investment advice, and the research method, authoring and copyright terms follow.
In July, Microsoft said something it had never been able to say before. Across its global operations in FY25 it put back more water than it took out, 14.28 million cubic meters replenished against 13.27 million withdrawn. The ledger crossed. That is hard to do while building data centers at the pace the AI build-out demands. It rests on 98 funded projects across the company’s replenishment portfolio, and Microsoft is entitled to say so out loud.
Report In Two MinutesReport synopsis · AI drafted · Analyst reviewed
Drafted by AI from the full research report and reviewed by a Greyhound Research analyst before publication. The analysis and conclusions are Greyhound Research’s own.
On September 29, 2026 Oracle announced Fusion Claw, a governed runtime that lets its Fusion agentic applications plan and execute work on their own inside the system of record, under authority the customer delegates. Oracle frames it around three constructs: an Enterprise Operating Envelope for the boundaries, an Outcome Trust Harness for each run, and an Outcome Receipt for the record. Greyhound Research tested the announcement against the four verbs it set out in July: name the identity, read the log, bound the authority, and reverse the action.
The documented record is real and dated: a per-application switch in Release 26D, fifteen named capabilities, a Fusion Claw row in Oracle’s contract that meters planning on frontier models, budgets with auto-stop, and a sentence stating that the AI cannot grant itself authority to act. Oracle has also confirmed in writing that deterministic execution draws no AI Units, that outcomes run asynchronously with no 60-second limit, that reasoning runs where the model provider offers the model, and that Fusion Claw depends on no OpenClaw component. The runtime itself, the receipt’s export and retention terms, units per outcome and a named customer are not yet documented. Greyhound Research judges bound the authority documented and strengthened, read the log the richest per-run record seen from a system-of-record vendor but conditional on a specification, name the identity bounded by the running user’s access, and reverse the action left with the system of record, as it is across the category.
For technology leaders the work is procedural before it is strategic. Set the boundaries in the roles, approvals and policies Fusion already holds, put the region of reasoning in writing, classify each outcome by whether it can be reversed, read one receipt against the audit tables in research mode, and measure units per outcome before the budget is set. The research report sets out the evidence, the Greyhound CXO Runbook, and the ten questions only Oracle can settle.
In This Research ReportSection guide · AI drafted · Analyst reviewed
Seventeen sections, each with a short summary. Select a title to go straight to that section.
Oracle launched Fusion Claw on September 29, a runtime that lets Fusion’s agentic applications plan and complete work under authority the customer delegates. The four tests Greyhound Research set in July are the measure it faces, and the terms are explained in plain words.
Claw is aimed at data-heavy, constraint-rich work such as rostering and freight consolidation. The customer chooses one of three autonomy levels, and the authority comes from roles, approvals and policies Fusion already holds.
Three Oracle documents name Fusion Claw: a per-application switch, fifteen named capabilities and a metered row in the contract. Availability follows Release 26D cohort by cohort, and no customer is named yet.
Oracle describes a container per outcome in which the planning model never touches the ledger. The design is the right one, but no Oracle document describes it yet, so buyers should treat it as a vendor statement.
Bound the authority is documented and strengthened, read the log is rich but unspecified, and name the identity is borrowed from the running user. The receipt does not yet name the policy or the person who delegated it.
Nothing in the Claw constructs unwinds a posted outcome. Reversal stays with Fusion’s own compensating transactions, as it does across the category, and the documented stop halts the whole fleet rather than one outcome.
Every Claw outcome plans on a frontier model that runs where its provider offers it, which for Mumbai, Singapore or Sydney means another continent. No control yet pins the reasoning to a region.
Regulators in seven jurisdictions converge on a four-part rule: answer for the agent, halt it, reconstruct what it did, and let a human change the result. The receipt meets the two parts about the record, not the two about intervention.
Oracle has confirmed in writing that Fusion Claw is its own architecture with no OpenClaw dependency. An equipment maker holds a live trademark application for the same name, so contracts should name the capability by function.
Buyers ask who answers for an autonomous outcome before they ask what it can do. Finance and HR are cautious, operations is further along, and buyers are ahead of vendors on agent identity.
Oracle’s contract changed on announcement day and meters a Claw action at 20x on a Frontier model. The design is sound for cost, but units per outcome, the number a budget needs, is unpublished.
The controls are documented, the runtime is claimed, and eight items are owed. The right response is staged trust: research mode first, then full auto one outcome class at a time.
Five actions, each with an owner: set the envelope and the unit cap, put the region of reasoning in writing, sort outcomes by whether they can be reversed, read a receipt against the audit tables, and measure units per outcome.
Ten questions only Oracle can answer, from the 25 applications by name to the receipt specification, a cancel for a single outcome and units per outcome. Greyhound Research will publish the answers as given.
Fusion Claw is new machinery, not new vocabulary, and confidence rose on two verbs and held on two. Oracle has built the part that obeys. The part that delegates is the buyer’s, set in controls Fusion already holds.
What Oracle Launched: A Runtime That Acts On Delegated Authority
On September 29, 2026 Oracle announced Oracle Fusion Claw, which its release describes as “a governed agentic execution runtime for Oracle Fusion Agentic Applications that combines AI reasoning with deterministic enterprise computation to enable highly complex work to be economically completed at scale.” Three constructs govern it in Oracle’s account: an Enterprise Operating Envelope, an Outcome Trust Harness, and an Outcome Receipt. For a technology leader the announcement describes something more specific than a runtime. It describes software that will act on its own inside the system of record, under an authority the buyer has to write down first.
Research Note In Two MinutesAI 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.
On August 31, 2026 the boards of ITC Infotech and Happiest Minds Technologies approved a combination of two of India’s mid-tier technology services firms. The transaction is announced and board-approved, not completed. Approvals are still to come, the companies operate independently until they arrive, and the window runs to roughly fifteen months.
Read as capability, this is two written-down gaps closing at once, a product engineering and generative AI gap at ITC Infotech and an enterprise applications gap at Happiest Minds. Each side is buying the half of the stack it could not build in time.
Happiest Minds earns about 80 percent of its revenue from product and digital engineering and brings its own platforms, Arttha among them. ITC Infotech brings the run side, more than 2,700 SAP consultants, more than 700 PLM professionals, and a managed services franchise weighted toward Europe. Together they carry more than 19,000 people and over 800 customers, against a stated ambition of USD 1 billion in revenue by FY28.
Greyhound Research’s position is that the consequential change is not the transaction but the redrawing of the middle of the Indian IT services market, where most enterprise buyers hold their leverage. The capability case is real on both sides. The risk sits in the integration seams, which surface in delivery rather than in announcements.
Four seams decide the outcome. Two agent platforms have to become one. Two security operations centers and two AIOps estates have to converge. The AI-first claim has to survive a customer’s governance review. And the engineers who make Happiest Minds distinctive have to stay.
For a technology leader with either firm on the vendor list, the supplier has changed shape and last year’s account plan no longer describes it. Greyhound Research recommends three asks, this quarter rather than next year: the integration plan in writing rather than the synergy narrative, who will run the combined engineering estate, and continuity commitments on the roles that hold the seams together. The research note sets out which management claims a buyer can verify and which they cannot, along with the triggers that would change this position.
On August 31, 2026 the boards of ITC Infotech and Happiest Minds Technologies approved a combination that the transaction materials describe in arithmetic: more than 19,000 people, over 800 customers, and a stated ambition of USD 1 billion in revenue by FY28. It is also the second business ITC has moved toward a public market in nineteen months, after the ITC Hotels demerger of January 2025 and on close to identical stated reasoning, which makes this listing a question of when and not whether. For a technology leader the same announcement describes something more specific.
On July 14, Oracle announced an AI-native builder experience for Oracle AI Agent Studio for Fusion Applications, spanning no-code, low-code, and pro-code development at what Oracle states is no additional cost to Fusion customers. Per Oracle, the experience would enable customers and partners to create and run Fusion Agentic Applications, a class of outcome-driven applications executed by coordinated teams of specialized AI agents, natively inside Oracle Fusion Cloud Applications.
Greyhound Research sorts this announcement against what Oracle AI Agent Studio for Fusion Applications already offers, and the sort yields four grades: the new, the extended, the expanded, and the established.
New: the AI Studio Skill. Per the release, the Skill anchors the pro-code path, bringing Visual Studio Code, Git-based command-line workflows, and AI coding assistants including Codex and Claude Code into Fusion-native development. A skill packages the platform knowledge that guides coding assistants.
In Greyhound Research’s view, Oracle’s intent is to bring agentic development closer to a conventional software lifecycle, with local validation and pipeline-based release replacing configurations trapped in an administration console; whether the resulting artifacts prove portable, reversible, and consistently governed across both build paths remains to be demonstrated.
Extended: the ecosystem, via GitHub. A promised public GitHub repository is slated to carry templates, starter projects, and reference architectures, together with sample applications and reusable assets, giving partners and developers a common starting point that Oracle’s studio tooling has not offered before.
Our read of the intent: Oracle is seeding a developer ecosystem around Fusion rather than a studio alone, because public code is how a platform recruits builders; the seriousness of that ambition will be measured by whether the repository arrives live, populated, and actively maintained over time.
Expanded: a marketplace of applications. The AI Agent Marketplace is set to grow from a catalog of individual agents to one of complete agentic applications, sitting alongside the existing portfolio of Oracle, partner, and third-party agents customers can already deploy inside Fusion.
Greyhound Research reads the intent as distribution at scale, turning partner-built applications into catalog products rather than one-off implementations; an application that installs from a catalog and runs as a governed Fusion artifact can travel through Oracle’s channel and partner network in a way bespoke builds never could.
Established: the studio and portfolio. Oracle AI Agent Studio launched in 2025; the natural-language and low-code builders, the approval workflows, and the interoperability protocols are documented capabilities, as are testing and observability, while Oracle’s own counts put the current portfolio at 22 Fusion Agentic Applications, 1,000-plus embedded agents, and 80,000 certified experts trained in the studio.
The intent, as we read it, is continuity: this release does not introduce its agentic direction; it attempts to turn that direction into a broader development and distribution model, built on capabilities Oracle already ships. The centre of gravity shifts from building or customising agents to building and running governed applications natively inside Fusion.
Table 1 grades the announcement in one view.
Greyhound Standpoint: At Greyhound Research, we believe the significance of this announcement is not more agents; the market is already saturated with those. The enterprise agent race has quietly changed shape, and the contest is no longer over who builds the smartest agent but over who owns the governed runtime in which agents are permitted to act. Oracle’s architectural claim is that agents born inside the system of record inherit the security, approvals, and auditability that agents built outside it must reconstruct at great expense.
Why This Matters To CIOs
Greyhound Fieldnotes, drawn from Greyhound Research’s advisory work, confirm that the problem Oracle is aiming at is the real one. Across our 2026 advisory work, we have repeatedly found that the hardest problem in enterprise AI is not building an agent; it is operating one. When agents are built outside the enterprise system, the organization must separately solve identity, approvals, and audits, along with the observability and lifecycle plumbing beneath them, and that is precisely where pilots stall.
Greyhound Research believes Oracle’s claim is credible because it attacks the hardest problem in enterprise AI. On detail, spanning identity semantics, logging depth, and reversibility, it remains unproven. Whether it holds will be decided by the technical documentation Oracle publishes, not by the announcement.
Greyhound Fieldnotes record a consistent pattern: the demonstration wins the room, and the deployment then dies in the corridor between the innovation team and the controller’s office. The agent may perform the task, but production stalls because machine actors were never provisioned within the identity model, the approval matrix, or the audit trail. That corridor is the difference between real governance and dashboard governance.
Greyhound CIO Pulse 2026, across more than 1,000 technology leaders, finds fewer than one in three can point to measurable outcomes across most of their funded AI portfolio. Adoption is not value. Pilots are not production. In agentic systems, the distance between the two is filled by identity, controls, auditability, and the uncomfortable question of who remains accountable when software begins to act.
Every major vendor is racing to own the answer, each positioning from its own center of gravity. Microsoft is building outward from productivity and the Power Platform, with an administrative control plane spanning its estate. Salesforce is anchoring on customer data, wrapping agent actions in its trust layer and CRM context. ServiceNow is treating the workflow itself as the governing structure, extending its control-tower discipline to agents. SAP is grounding agents in harmonized processes and business-data context across its suite, while Workday is positioning the agent as a governed object inside its system of record for people and money. Beyond the suites sit two further archetypes: independent orchestration platforms, which offer breadth across models and tools yet must reconstruct enterprise controls from the outside, and automation-led alternatives, which execute reliably within scripted boundaries and reason poorly beyond them.
Greyhound Standpoint: At Greyhound Research, we believe Oracle is making the same bet as its rivals, just deeper. Oracle is betting on the depth of Fusion-native business objects, transactional workflows, and inherited application controls as the foundation for agentic execution; the argument is that the control plane belongs inside the application runtime because that is where the approval chains and audit trails already live. If that claim holds, the payoff is practical: a company can move from AI that recommends an action to AI that takes the action itself, under the controls it already trusts, without first building the governance plumbing from scratch.
What Oracle Must Still Prove
Greyhound Research’s position is that five questions decide whether the governance claim survives contact with a real-world customer deployment.
Identity. Does each agent operate as a distinct principal under least privilege, or does it borrow the entitlements of a human user?
The record. What is logged, from prompts and tool calls to the model version behind each action, and can the customer export it?
Reversibility. A bad chain of actions must be unwound through compensating transactions once it has executed; what mechanism exists, and who triggers it?
Model governance. Oracle documents customer choice across its own curated models and external ones, which makes substitution a routine event; a swap beneath a live application needs to be approved, traced, and benchmarked.
Third-party participation. How are external agents arriving over the MCP and A2A interoperability protocols authenticated, sandboxed, and revoked?
The marketplace adds a sixth question. Oracle publishes a 21-point enterprise readiness checklist for partner-built agents, an unusually concrete disclosure by the standards of this market, yet the harder questions begin after inspection: who carries liability when a partner-built application misfires inside a customer’s ledger, how updates are controlled once that application is live, and what commercial terms bind Oracle, the partner, and the customer when something breaks.
Lest we forget, certification is a gate and accountability is a contract.
A Word Of Caution For CIOs
Greyhound Research cautions that a native runtime relocates complexity; it does not cancel it. What such a runtime genuinely removes is the bolt-on burden: security, approvals, and audits can be reused rather than rebuilt around the agent because the application platform already carries them. What remains moves to harder ground, domain by domain, and our work with buying organizations shows the year-one bill takes a different shape in every estate, with none of the shapes zero.
Employee services run light; employment decisions run heavy. Bounded services such as policy guidance and scheduling carry the lightest lift, because authority stays narrow and approvals explicit. Employment, compensation, and promotion decisions sit at the other end, touching identifiable human beings and converting software defects into legal and reputational events; they fall in the EU AI Act’s high-risk category for employment under Annex III, whose enforcement the newly adopted Digital Omnibus defers to December 2, 2027, while anti-discrimination and data protection laws bind them today. The map is wider than Brussels: the United States runs a patchwork of state, sectoral, and consumer-protection enforcement rather than one omnibus regime, and India’s Digital Personal Data Protection framework, its rules notified in November 2025, is moving through phased implementation to full enforcement by 13 May 2027, which puts personal-data duties on agentic deployments now, not at some future enforcement date.
Finance’s value comes with priced-in controls and audits. The heaviest controls burden sits here: the value in close acceleration and collections is real, and so is the cost of controls engineering, audit sign-off, and process redesign, because a controller will not accept a trace that cannot be walked. Reversibility matters most in this domain, since a single bad chain of postings must be unwound through compensating transactions, with a named owner for the trigger, before the auditors arrive.
Customer errors are intolerable but visible; pilot here. The bill arrives in integration and override design, since the tolerance for a wrong promise made to a paying customer is close to nil. The compensating advantage is visibility: unlike supply-chain errors, a customer-facing mistake surfaces immediately and hands to a human naturally, which is why service escalations sit among the very first sensible pilots, provided the override path is designed well before the agent ever speaks to a customer.
Stale records cap autonomy; cleanup comes before automation. Master data quality becomes the ceiling on autonomy, because an agent acting at machine speed compounds a single stale supplier record into a cascade of confident, well-logged mistakes, and the physical world reports them late. The first bill is therefore data remediation, and the sequencing follows from it: exception handling belongs strictly after the cleanup, not before, because autonomy built on stale records automates the error rather than the business process.
The riskiest failures hide at the cross-suite seams. Almost no enterprise is a single-suite estate, so real work crosses runtime boundaries as a matter of course, and at each crossing the agent’s identity, its authority and its audit trail change jurisdiction. Drift at those seams combines the highest impact with the lowest detectability of any failure mode in this category, and in the largest estates, partner fees, governance boards, and change management quickly overtake the software line itself.
Zero Oracle cost is not zero customer cost. Oracle states the capabilities arrive at no additional cost to Fusion customers, and the subscription genuinely bundles the studio. Oracle’s own marketplace documentation is explicit that agents deployed from partner-built templates count as custom AI and carry a separate subscription fee. The full year-one bill still arrives through implementation and controls engineering, partner and integrator fees, and infrastructure consumption at scale; the bundling decision therefore changes which budget carries the spend rather than eliminating the spend itself. The boundary between them is a contracting question, taken up in the engagement conversations below.
One caution outlasts the list: the AI Act deferral moves the enforcement date without moving the accountability. And one question spans every row above: whose control plane governs the agent whose task crosses three vendors’ runtimes? No vendor racing for this category has answered, and the buyer who asks first negotiates from strength.
How CIOs Should Engage Oracle: Six Conversations
Greyhound Research advises Fusion customers weighing Oracle AI Agent Studio and Fusion Agentic Applications to structure the Oracle engagement around six conversations, each of which can begin with the account team this quarter.
Scope the first Oracle pilot. Ask the account team to help stand up a use case that is bounded, measurable, and reversible, where governance inherits from day one. Collections and service escalations meet those tests today; hold the financial close to advisory or exception-led mode until transaction traceability and compensating controls are demonstrated, and let supply chain exception handling wait for clean master data. A pilot chosen on these criteria produces evidence; a pilot chosen on enthusiasm produces a demo.
Put five questions in writing. Ask the account manager for written answers to the questions this note itemizes: whether each agent operates as a distinct principal under least privilege or borrows a human’s entitlements; what is logged, from prompts and tool calls to the model version behind each action, and whether you can export it; what compensating-transaction mechanism unwinds a bad chain of actions and who triggers it; how a model swap beneath a live application is approved, traced, and benchmarked; and how external agents arriving over MCP and A2A are authenticated, sandboxed, and revoked. Add the cross-suite question: Which control plane is the plane of record when an agent’s task crosses into your Salesforce, Workday, or ServiceNow estates. Any platform that asks to act inside your ledger owes the same answers, so the list travels to every suite vendor, including those yet to announce.
Demand AI Studio Skill specifics. If your developers will build rather than configure, ask for the Skill’s documentation, access to the promised repository, and a working demonstration before committing engineering time: how a Fusion Agentic Application is represented as an artifact under Git, how the same artifact behaves across the no-code and pro-code paths, and how local validation and pipeline-based release interact with Fusion’s runtime governance. The documentation should also show versioning, rollback semantics, agent identity boundaries, approval threshold configuration, transaction integrity across multi-step actions, and reasoning trace access. Whether those artifacts prove portable, reversible, and consistently governed is the announcement’s central open question; have Oracle answer it in your environment, not in a keynote.
Define what the bundle covers. The subscription bundles the studio; the year-one bill still arrives through implementation and controls engineering, partner and integrator fees, and infrastructure consumption at scale. Ask which of the 22 Fusion Agentic Applications and the 1,000-plus embedded agents your existing subscription already includes, which arrive with partner commercials attached, and which budget carries each line, in writing, before deployment rather than after.
Settle partner-application liability before installing. The marketplace questions raised above become a contracting conversation here: before installing a partner-built application, get written answers on who carries liability when it misfires inside your ledger, how its updates are controlled once live, and which commercial terms bind Oracle, the partner, and you when something breaks. Inspection is Oracle’s gate; the accountability contract is yours to negotiate, and it is negotiated before installation, not at the incident review.
Negotiate the exit before entry. Native governance is real, and it is also non-portable. Every agentic application built as a Fusion runtime artifact deepens platform gravity, so a buyer exchanging model-vendor optionality for application-platform dependency should arrive at the signature with exit clauses drafted and portability already tested. Ask the account team to demonstrate export: what leaves with you, in what format, and what remains behind.
None of these conversations require waiting for Oracle’s documentation, for a regulator, or for the market to settle. Each converts a buyer from a spectator of this contest into a participant who sets terms, and together they prepare an enterprise for the larger transition this announcement belongs to.
The Final Greyhound Research Standpoint
Greyhound Research holds that enterprise software is moving from recording work to executing it, and the transition will be won by the vendor that makes autonomy boringly governable. “Boring” is not a small word in this market; it is the highest compliment an enterprise system can earn, because “boring” means predictable, inspectable, and reversible, and those are the qualities a controller, an auditor, and a regulator will sign their names against.
The test is simple, and it travels well beyond Oracle.
Name the identity. Every agent runs as its own principal, holding the least privilege its task allows, never borrowing a human’s credentials.
Read the log. Every prompt, tool call, approval, and model version behind an action can be inspected and exported by the customer, not only by the vendor.
Bound the authority. Approval thresholds are set by the enterprise as policy on the transactions that matter and cannot be quietly widened.
Reverse the action. A bad chain of decisions can be unwound through compensating transactions, with a named owner for the trigger.
A buyer who cannot do these four things does not have an agentic application; it has an ungoverned employee with system access.
Oracle starts from a position of genuine strength. Against that test on day one, the public record is already strong where it matters first: agents acting on native business objects, under approval thresholds the enterprise sets. The remaining verbs, the full log, a distinct agent identity, and a demonstrated reversal remain open proof points: the public record neither establishes them nor establishes their absence, and the documentation will settle it. Across the leading vendors’ public material reviewed for this note, none yet documents all four in full, so the gap belongs to the category, not to Oracle. What it measures is how young this category still is and how much of its story will be written in documentation rather than announcements.
For Oracle, the path from here is concrete and entirely within its control: publish the technical documentation, bring the promised repository live, put named customers into production with measured outcomes, produce external audit artefacts, and give buyers the pricing and audit clarity they can carry into their own governance processes. Each of those steps converts the announcement’s promise into proof, and Oracle brings credible assets to the path: Fusion’s transactional depth, an established partner network, and its certified training base. Greyhound Research will track that evidence as it lands, and a full examination of the governed-runtime category will follow. Over three years, the market will split between platforms that can prove governed action inside systems of record and platforms that remain glorified orchestration layers.
Greyhound Standpoint: At Greyhound Research, we believe the prize in this market is accountable autonomy: agents that act under the controls an enterprise already trusts, with a named owner when they act wrongly. Oracle has moved the contest onto the ground where that prize is decided, and it has done so with the machinery of a runtime rather than the language of a demonstration. The evidence will now do the talking. Everything else is a demonstration.
Important Disclaimer
Greyhound Research are industry analysts, not equity analysts. This note is a demand-side assessment of Oracle’s announcement, architecture, and buyer implications; it is not investment advice, and nothing in it should be read as a view on securities. This note draws on pre-briefing materials provided under embargo by Oracle and on Greyhound Research’s own analysis; the announcement became public on July 14, 2026, and key figures were re-verified against public sources on July 19, 2026.
This is a strictly independent analysis: Oracle has not commissioned, paid for, reviewed, or influenced this note in any form, and Greyhound Research has received no compensation from Oracle or any other party for its production. The views expressed are Greyhound Research’s alone.
This note draws on Greyhound CIO Pulse 2026 and Greyhound Fieldnotes. Our full archive is available at greyhoundresearch.com.
Sources And References
Every load-bearing claim in this note traces to one of the sources below, presented in Harvard reference format and grouped by class. In-text author-date citations are deliberately omitted, as they serve academic writing rather than research intended for the executive desk. All sources accessed 19 July 2026.
Greyhound Research (2026a) Greyhound CIO Pulse 2026. Proprietary survey of more than 1,000 enterprise technology leaders.
Greyhound Research (2026b) Greyhound Fieldnotes. Advisory engagements with enterprise technology buyers, 2026.
Analyst In Focus: Sanchit Vir Gogia
Sanchit Vir Gogia, or SVG as he is popularly known, is a globally recognised technology analyst, innovation strategist, digital consultant and board advisor. SVG is the Chief Analyst, Founder & CEO of Greyhound Research, 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. SVG is also the Founder & CEO of The House Of Greyhound, an eclectic venture focusing on interdisciplinary innovation.
Copyright Policy. All content contained 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. We request our readers not to copy Greyhound Research content and not republish or redistribute them (in whole or partially) via emails or republishing them in any media, including websites, newsletters, or intranets. We understand that you may want to share this content with others, so we’ve added tools under each content piece that allow you to share the content. If you have any questions, please get in touch with our Community Relations Team at connect@thofgr.com.
A Global, Award-Winning, Technology Research, Advisory, Consulting & Education Firm
How We Help
Click on the picture to read more.
Greyhound Research is the trusted source of insights and expert advice for over 200,000 professionals and 500+ companies globally.
We serve every player in the technology ecosystem, from buy-side CxOs and sell-side vendors to venture capitalists, private equity and fund managers, startups and growth companies, and government and industry organizations.
For tailored insights that truly move the needle for your organization, we’re the partner you need. Our Human+AI approach combines our deep industry expertise with AI’s transformative power to deliver research, advisory, consulting and education engagements tailored to your business.
Inspired by the greyhound, our brand mirrors its qualities: swift, agile, flexible, and independent. We excel at uncovering crucial insights that drive innovative, impactful decisions.
Greyhound Research insights are read across all continents. (Source: Website stats via WordPress & Google Analytics) Click on the picture to read more.
Ready to fuel your organization’s next breakthrough? Contact our Client Centricity Team today.
AI isn’t replacing research—it’s enhancing it. Our Human+AI approach combines human-led critical thinking with AI-led deep analysis, delivering nuanced insights that AI alone can’t match.
Greyhound Research equips companies and executives with robust data, innovative tools, actionable insights, expert advice, and critical skills to lead transformation in these changing times.
We take a tailored, high-impact approach, combining research products, advisory services, consulting, and education to deliver real, actionable insights. Instead of a one-size-fits-all model, we blend these offerings to meet each client’s unique needs, ensuring they get the right mix of data, expertise, and hands-on guidance.
This integrated approach means clients don’t just get reports—they get strategic insights, expert advice, and practical education that drive real outcomes. Here’s how we make it happen.
Click on the picture to read more.
Get insights directly in your inbox. Subscribe to Greyhound Digital Digest to receive expert analysis.
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 Oracle AI…
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 Oracle AI…
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 Oracle AI…
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 Oracle AI…
A practical guide by Greyhound Consilium, in partnership with Oracle 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…
Report In Two MinutesReport synopsis · AI drafted · Analyst reviewed Drafted by AI from the full research report and reviewed by a Greyhound Research analyst before publication. The analysis and conclusions are Greyhound Research’s own. On October 8, 2026,…
Report In Two MinutesReport synopsis · AI drafted · Analyst reviewed Drafted by AI from the full research report and reviewed by a Greyhound Research analyst before publication. The analysis and conclusions are Greyhound Research’s own. In July 2026 Microsoft…
Report In Two MinutesReport synopsis · AI drafted · Analyst reviewed Drafted by AI from the full research report and reviewed by a Greyhound Research analyst before publication. The analysis and conclusions are Greyhound Research’s own. On September 29, 2026…
Research Note In Two MinutesAI 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. On August 31, 2026 the boards…
On July 14, Oracle announced an AI-native builder experience for Oracle AI Agent Studio for Fusion Applications, spanning no-code, low-code, and pro-code development at what Oracle states is no additional cost to Fusion customers. Per Oracle, the experience would enable…
On July 14, IBM released preliminary second-quarter results a full week ahead of schedule. Revenue up 1 per cent. Software up 5 per cent. Infrastructure down 7 per cent as the strongest mainframe launch in the company’s history wrapped, and…
ServiceNow on Tuesday announced that it would buy cybersecurity vendor Armis for $7.75 billion in cash. This builds on its December purchase of identity security vendor Veza, and the closing of its acquisition of AI vendor Moveworks. Sanchit Vir Gogia,…
There are seasons in enterprise technology when the landscape shifts quietly. Then there are moments like this one when the ground moves in full view of the industry. IBM’s decision to acquire Confluent sits in the latter category. It marks…
For decades, Power servers stood as quiet enablers of core enterprise operations, rarely questioned, rarely credited. But with Power11, IBM isn’t nudging the platform forward. It’s recasting the role of infrastructure itself. This isn’t a performance leap. It’s an operational…
IBM closed Q3 2025 with 16.3 billion dollars in revenue, up 7 percent in constant currency and 9 percent year over year, its fastest pace in several years. All three core segments grew sequentially, and operating discipline showed up clearly…
In recent weeks, three of the world’s most powerful AI platforms – OpenAI (ChatGPT Go), Google (Gemini), and Perplexity – quietly dropped a bombshell in India. They made their premium AI services free. To the public, it feels like a…
Prefer watching instead of reading? Watch the video here. Prefer reading instead? Scroll down for the full text. Prefer listening instead? Scroll up for the audio player. Infosys entered Q2 FY26, crossing the five-billion-dollar revenue mark, up 2.9 percent year…
Here’s a selection of testimonials from our community. To read them all, click here.
Love at first sight. Had heard about it, seen it in movies and read in novels but felt it on my day one at Greyhound. The very talented go-getters of Greyhound, the cozy office and the entire feel of it fills you up with all the positivity and energy. Although my span with them was extremely short, but every time I hear about their achievements it fills me up with a sense of pride and joy and my love for Greyhound grows multifold. Thank you all for such wonderful memories…..Wish Greyhound many more success stories…you guys Rock!!!
“I’m honoured to join Greyhound Research Advisory Board and looking forward to an exciting journey with Sanchit Vir Gogia and his team as they evolve their strategy focused on clients in business and talent transformation. Greyhound and Sanchit challenge conventional thinking to digital innovation, which is refreshing and provides their clients with unique perspectives and better outcomes. I’m truly looking forward to contributing to Greyhound and being part of an esteemed group of industry specialists.”
“It gives me immense pleasure to be a part of the Advisory Board of Greyhound Research as an Honorary Advisor. The thing I like most about Greyhound Research is that they always have an unbiased and unconventional view on technology and technology partners when it comes to evaluation. As organizations embark on their digital transformation journey, there is a lot that organisations and their leaders can benefit from Greyhound Research. I look forward to working closely with my fellow advisory board members to take Greyhound Research to newer heights.”
“Research needs to be fundamentally need-driven and without the bias of other unfitting comparatives. There is a genuine need for a grounds-up research organisation which understands the lay of the land. I see Greyhound as a huge sincere try in that direction. I wish to be a part of their journey.”
“I am delighted to join the advisory board of Greyhound Research with a simple view to help its customers be more successful. Having spent close to 25 plus years in ICT across APJ, I wanted to share my insights and learnings to enterprise’s seeking help. I am hopeful and confident that my experience would add value to Greyhound and its esteemed customers.”
Thank you for taking out time to be part of Dimension Data’s Leadership Connect Event Series. Thanks to your insights and clarity, the feedback from our CIO/CTO clients have been absolutely fantastic. I and my global leadership team appreciate the way you helped us de-mystify Digital Transformation and articulate why Hybrid IT is key for organisations in this Digital era. The research produced by you folks at Greyhound Research is highly objective and does great justice to demystify the hyped technology jargons. What truly stands out about your work is your effort to make Digital Transformation relevant for organisations across countries and verticals. Kudos to the team!
We have had the pleasure of working with Sanchit and his team at Greyhound for close to 2 years now and the first thing which strikes you about his incredibly talented team is their approach to research and analytics. They are flexible yet ethical, agile with process and aligned with business objectives with an unbiased commentary. Their focus on customer needs and demands with sharp insights on industry, make them a formidable voice in the analyst community. They differentiate themselves with honour and align themselves with pride. It has been a pleasure to work with Greyhound and we look forward to more opportunities in the future.
This answer was generated by an AI system. It was not written by Greyhound Research. It is produced automatically from Greyhound Research's published articles and reports, and is meant to help you find the relevant material quickly.
Because it is generated, it may simplify, condense or leave out parts of the original analysis, and it may phrase things differently from how Greyhound Research has stated them. Any statistics, percentages, quotations or named references in this summary may not match the source exactly, and should be checked against the original articles linked below before you rely on them or cite them anywhere. The wording can also change from one search to the next.
This summary is not an official Greyhound Research position, recommendation or advice. It should not be quoted or attributed to Greyhound Research as a statement of its views, and it is not a substitute for reading the original analysis in full. For the authoritative version, please follow the source links to the original pieces.
For guidance on your own specific context, write to us at connect@thofgr.com.
Greyhound Research
Greyhound Research is a Global, Award-Winning, Technology Research, Advisory, Consulting & Education Firm
