Fusion Claw at a glance
- What it is
- Oracle calls it a "governed agentic execution runtime" for Oracle Fusion Agentic Applications. Put simply, it is the engine those applications run on when the work is large, long or complex.
- Announced
- 29 September 2026
- Launch applications
- 25 Claw-powered applications in finance, HR, supply chain and sales. That takes Oracle's Fusion Agentic Applications to 75 in total.
- AI models
- Gemini and OpenAI models at launch. Oracle says it plans to support more frontier models over time.
- Runs on
- Oracle Cloud Infrastructure (OCI)
- Cost
- Trade press reports that Claw comes at no extra charge with the Fusion Agentic Applications offering, which is bought separately from your standard Fusion subscription. Usage is paid for in Oracle "AI units". Confirm your own terms with Oracle.
- Availability
- Oracle's announcement says the 25 applications are available now. Some reports describe a rollout continuing into October 2026, so check what is actually enabled in your own environment.
Why Oracle built it
Most AI in business software so far behaves like a helpful colleague you ask a question. It reads, it summarises, it recommends. Then a person still has to do the work.
That works for small tasks. It breaks down for big ones. Reconciling thousands of journal entries, rebuilding a hospital staff roster, or comparing hundreds of ways to combine shipments is too much for a chat answer. It is also too expensive and too risky to let an AI model "guess" its way through every calculation.
Fusion Claw is Oracle's answer to that gap. It moves Fusion's AI from assisting with work to doing the work, inside limits the business controls.
Think of Fusion Claw as a new skilled worker
The easiest way to understand Claw is to picture hiring a very capable new team member. Every part of Claw has a familiar workplace equivalent.
| A good human worker has… | In Fusion Claw this is… | What it means |
|---|---|---|
| A brain that plans and adapts | A frontier AI model (Gemini or OpenAI) | Understands the goal, makes a plan, and changes the plan when something changes. |
| A calculator, spreadsheets and the company systems | Deterministic enterprise computation | Exact calculations, business rules, optimisation and transactions. Same input, same answer, every time. |
| A job description and a delegation-of-authority letter | The Enterprise Operating Envelope | Your objectives, procedures, policies, permissions, risk limits, who decides what, approvals and when to escalate. |
| A security badge that only opens certain doors | The Outcome Trust Harness | For each job, it controls which identity, data, capabilities and actions are allowed. |
| A private desk for the task | An isolated workspace per job (as reported by trade press) | Each job reportedly runs in its own sealed workspace that is removed when the job ends. |
| A work report handed to the manager | The Outcome Receipt | A record of the authority used, the evidence, the decisions, the transactions and the result. |
| A manager who decides how much freedom to give | The automation level you choose | From quick assistance, to "show me the plan first", to full automatic execution within delegated authority. |
The key idea is the split between thinking and doing. A good accountant does not "estimate" a total; they plan the approach, then use a calculator. Claw works the same way. Oracle's own framing is that AI reasoning is used only where judgement is needed, and cheaper exact computation handles the volume.
How one Claw job runs, step by step
- You set the goal. For example: "reconcile this period's accounts and explain every exception."
- The rules are already in place. The Enterprise Operating Envelope says what this job is allowed to touch, what limits apply, and what needs a human approval.
- The AI model makes a plan. It breaks the goal into steps and decides what to check first.
- Exact software does the heavy lifting. Calculations, matching, simulations and transactions run as normal deterministic code, not as AI guesses.
- The plan adapts. If a result is surprising, or something changes (a person calls in sick, a truck is full), the AI model re-plans. Oracle calls this continuous re-planning.
- A human steps in where you said they must. Depending on the automation level, a person reviews the plan before anything is posted, or Claw executes within its delegated authority.
- You get an Outcome Receipt. What authority was applied, what evidence was used, what was decided, what was posted, and the result. This is what an auditor or process owner reviews.
Want the technical picture? See the Fusion Claw architecture diagram in our AI Agents section.
The four launch examples Oracle named
Oracle launched 25 Claw-powered applications and highlighted four of them. These are the ones we can describe with confidence. The everyday scenarios under each are our own simple illustrations of the job Oracle describes.
OracleLedger (finance)
What Oracle says it does: reconciles accounting entries, investigates exceptions, and performs financial analysis to help close the books.
OracleWorkforce Staffing (HR)
What Oracle says it does: builds executable staffing plans that balance skills, accreditations, availability, schedules, labour rules, cost and service requirements.
OracleShipping Consolidation (supply chain)
What Oracle says it does: models consolidation alternatives across timing, capacity, service commitments and cost, to reduce shipping cost.
OracleAccount Territory Growth Plan (sales)
What Oracle says it does: models and optimises territory configurations and opportunity allocation against capacity and business constraints.
More use cases where this pattern fits
Oracle has not published the full list of 25 applications in its announcement. The examples below are not Oracle-announced applications. They are jobs that share the same shape: lots of data, clear rules, some judgement, and a need for an audit trail. Use them to spot candidates in your own processes, then check Oracle's application list for what actually exists.
| Area | Illustrative job | Why it suits a Claw-style worker |
|---|---|---|
| ExamplePayables | Investigate possible duplicate or mismatched supplier invoices before payment | High volume, clear matching rules, exceptions need explanation |
| ExampleProcurement | Find spend that could move to existing contracts or fewer suppliers | Needs analysis across many records, then a judgement |
| ExampleInventory | Explain why stock counts and system balances differ, and propose adjustments for approval | Rule-based checks plus investigation; approval before posting |
| ExampleCost accounting | Work through costing exceptions in period close and document each fix | Repeated every period; evidence matters for audit |
| ExampleHR | Rebalance shifts after leave requests while keeping certifications and labour rules | Constraint solving that changes daily |
| ExampleSales | Re-plan account coverage after a rep leaves mid-year | Scenario modelling with clear capacity limits |
What it means for each Fusion user
Finance teams
Less manual reconciliation, a shorter list of real exceptions, and an Outcome Receipt you can hand to audit. Your role shifts from doing the matching to reviewing the decisions.
HR and workforce managers
Staffing plans that respect skills, accreditations and labour rules, and that can be rebuilt quickly when people are absent.
Supply chain and procurement teams
Many more options compared before a decision, such as how to combine shipments, with cost and service trade-offs shown clearly.
Sales operations
Territory and coverage planning based on scenarios rather than spreadsheets passed around for weeks.
Functional consultants
A new kind of design work. You will help customers write the Operating Envelope: the policies, limits, approvals and escalation rules. Clean master data and well-defined processes become even more valuable, because Claw executes the process you define.
IT, security and audit
Claw's design puts governance at the centre: scoped permissions per job, isolated execution, and a receipt for every run. Your job is to decide what authority to delegate, and to review the receipts.
How Claw can raise output, and where the gains come from
- Scale. Large, long-running jobs are possible because the heavy work runs as exact software, not as AI conversation.
- Lower AI cost per job. The expensive AI model is used for planning and judgement, not for every calculation. That is Oracle's core economic argument for Claw.
- Speed of reaction. Continuous re-planning means a plan can be rebuilt when conditions change, instead of waiting for the next manual cycle.
- Consistency. Deterministic computation gives the same answer for the same input, which matters for finance and compliance.
- Trust and auditability. Each run leaves an Outcome Receipt, so a reviewer can see why something was done, not only what.
- People on higher-value work. Staff move from repetitive processing to reviewing exceptions and making decisions.
Oracle has not published customer results for Claw yet, so treat any specific time or cost saving as something to measure in your own pilot, not something to promise.
Limits and honest cautions
- It runs your rules, good or bad. Analysts point out that results depend on clean data, standard processes and clearly defined limits. Messy master data or vague policies will limit the value.
- Cost is usage-based. Claw jobs consume Oracle AI units. Info-Tech Research Group warned that actual costs vary with the actions performed, model choice, testing and human review, and that unused pooled units can expire. Read your contract terms.
- Start with review, not full auto. Oracle's Natalia Rachelson told SiliconANGLE that any task can run in full-auto mode, but customers are expected to begin with people reviewing plans. That is the sensible path.
- Model choice is limited at launch. Gemini and OpenAI models only, with more promised later.
- It is new. Independent customer evidence and audit outcomes are not published yet. Futurum Group analyst Keith Kirkpatrick told TechTarget that end-to-end automation is still "some ways away", because it depends on accurate data and complex coordination.
How to get started, in five steps
- Check access. Confirm with Oracle whether your organisation has the Fusion Agentic Applications offering and which Claw applications are enabled in your environment.
- Pick one painful, rule-heavy process. Reconciliation, scheduling or consolidation are good shapes. Avoid processes that change every week.
- Write the envelope before the pilot. Agree the objective, policies, risk thresholds, approval points and escalation path with the process owner, security and audit.
- Run in review mode first. Let people approve every plan. Compare Claw's results and receipts with the current manual process.
- Increase authority slowly. Only delegate more autonomy where the receipts show consistent, correct decisions. Track AI unit use alongside time saved.
Our Build Readiness Checklist covers the data, security and testing questions you should answer first.
Claw vs Agent Studio vs Agentic Applications vs OpenClaw
| Name | What it is | Simple analogy |
|---|---|---|
| AI Agent Studio | Oracle's toolkit for building and customising agents in Fusion | The workshop |
| Fusion Agentic Applications | Ready-made AI applications and workspaces Oracle ships, 75 in total after this launch | The finished products |
| Fusion Claw | The governed runtime that powers 25 of those applications for complex, long-running work | The engine inside the newest products |
| OpenClaw | A separate open-source personal AI agent project that became popular in 2026 | The idea that inspired the name |
Coverage of the launch reports that Fusion Claw is inspired by the "Claw" class of agents named after OpenClaw, but that Oracle built it from scratch on its own architecture. It is not a version of OpenClaw.
The bottom line
Fusion Claw is a real step from AI that advises to AI that executes, with governance built in rather than bolted on. For Fusion users, the opportunity is big: hand the heavy, rule-based work to a digital worker and keep people on the decisions. The winners will be the teams that prepare well, with clean data, clear rules, a careful pilot and an honest look at every receipt.
Checked Facts checked on 1 Oct 2026 against Oracle's announcement, Oracle Extends Fusion Agentic Applications with Introduction of Fusion Claw (29 Sep 2026), and launch coverage from SiliconANGLE, TechTarget, CIO, Futurum Group, ERP Today and The Letter Two. Pricing, the isolated workspace and the sick-call re-planning example come from trade-press reporting, not the announcement itself. Scenarios marked "Example" are our illustrations. Availability and licensing vary; verify in your own environment and contract.