Oracle Added Claude Code to Fusion. Its Analyst Ranking Didn't Move.

Oracle shipped a third layer of its Fusion agent platform on 29 September 2026, and the analyst tier that marks it as a niche player in agent platforms is still the one Futurum shows today.

The new layer is called Fusion Claw. It is a runtime that lets Oracle's agentic applications run longer, more autonomous jobs, and Oracle launched 25 applications built on it. That moves the pitch a long way from March, when the builder was a no-code tool for business users, and from July, when engineers could script it with Claude Code. A B2B SaaS team selling into Fusion accounts now has to ask a harder question: does the bundled option cover the job my product does?

Skip the press-release framing for a moment. The useful facts are what shipped, what it costs, and where independent analysts still place the vendor.

What changed between March and October

Four releases in seven months, each adding a layer. The table is the whole timeline, with dates taken from Oracle's own announcements.

Chris Leone, an Oracle applications executive, called the first version "the foundation for a more autonomous enterprise" when it launched at the company's London conference. Business users could compose agents and connect internal data without writing code. The April waves gave them ready-made starting points, named things like a claims-settlement workspace and a sourcing command center.

By mid-July engineers got their own door. They could work in Claude Code, OpenAI Codex, an editor or plain Git, and the finished agent still inherited the platform's security and audit controls. Leone's language shifted with it: "software is moving beyond systems that record work to systems that actively drive and execute outcomes."

What the runtime actually does

Oracle describes Claw as a governed execution runtime. A frontier model reasons, plans and adapts at the start of each job. Deterministic enterprise computation then does the high-volume execution. Splitting the two is the whole idea: pay for model reasoning only where judgement is needed, and run the repetitive arithmetic on ordinary compute.

The portfolio now stands at 75 agentic apps, including the 25 new ones, the company says. They run on Oracle's own cloud, powered by Gemini and OpenAI models, with more models planned. For scale, the current flagship record for Google's family is Gemini 3.5 Pro. The release gives four examples:

• A finance app that helps leaders close the books and clear accounting exceptions.
• A staffing app that builds schedules respecting skills, labor rules and coverage.
• A shipping app that models ways to consolidate freight and cut cost.
• A sales app that compares territory scenarios against capacity.

Control is the other half. Oracle says each run operates inside an "Enterprise Operating Envelope" of policies, permissions, risk thresholds and approval rules, and that every finished job produces an auditable "Outcome Receipt". Customers pick how much automation a process gets, from quick assistance up to governed full auto-execution inside delegated authority. The approval gates piece explains why that dial matters for any agent that acts on its own.

What it costs

The release names no price for the Claw applications. The March and July announcements said the builder came at no additional cost to existing Fusion subscribers, and we could not confirm that the new apps follow the same rule. Ask your Oracle contact before you quote a buyer's budget to anyone.

Oracle does make one economic claim: separating reasoning from computation gives customers "a potential economic advantage." That is a design argument, not a measured saving. No customer figure appears in the release, so treat the word "potential" as doing real work.

General-purpose agent builders publish prices you can check. The best AI agent builders guide lists them, the personal AI agent cost piece shows how fast usage-based bills move, and the OpenAI agents pricing breakdown shows what a rival stack charges.

Who this actually helps

The pro-code path and Claw both target a narrow reader: an engineering team already committed to this ecosystem that wants to script a specialist agent instead of clicking through a builder.

• Fusion customers whose finance, HR or supply chain processes map to one of the 75 named apps.
• Consulting partners, several of which supplied quotes for the launch.
• Teams that built compliance workarounds around outside data access, since an analyst at Constellation Research called the earlier release a way to keep "agents, security, APIs, access, and governance" in one stack.

If your product connects to Fusion data through a workflow tool like Make, a native route now exists for anything that needs tighter access than an outside integration allows. Test it on a low-stakes workflow before betting a roadmap on it.

Who it narrows out

It is a worse update for anyone selling a point solution into these accounts. A buyer evaluating an agent layer for HR or finance work now has a bundled option before looking anywhere else. The best AI assistant for office work ranking shows how crowded that shelf already is.

That includes builders aimed at smaller teams, like Lindy, which sells a no-code pitch to a very different buyer. It includes open-source frameworks such as CrewAI, the usual route for a custom agent outside a vendor's walls. And it includes search-led tools like Glean, competing for the same platform-first budget conversation.

None of them lose their reason to exist over one release. A team five systems deep on an open-source stack is not migrating for a runtime launch. A regulated buyer that already trusts one security model rarely switches over a quarter's feature list. Still, a bundled option inside a license already paid for is real competition for the budget those vendors were counting on.

The ranking that didn't move

The obvious reading of four releases in seven months is that Oracle is closing the gap with established leaders. Futurum Group's Signal report on agentic AI platforms for enterprise says otherwise. It scores ten vendors out of 100 and sorts them into four zones by how well each can act as a control plane for enterprise agents.

Futurum describes the Aspiring group as vendors with established niches in database management and automation that need broader horizontal interoperability to compete as control planes. The report was published on 1 June 2026, so it predates Fusion Claw by four months. The Futurum page read on 23 October still carries these placements.

The case against reading too much into the tier

A tier published before a product ships cannot score that product. Claw is aimed squarely at what the report grades: identity, policy, observability and governed execution. If Futurum rescored with Claw in hand, Oracle could move up a zone.

There is also a structural argument. Oracle runs the system of record for finance and HR at many of its customers. A control plane that lives inside that system does not need to cross a rival cloud to be useful to a buyer who runs everything in Fusion. The cross-stack axis that Futurum weights heavily matters most to customers with finance in one place and sales in another.

Treat the analyst endorsements with the same care. An IDC analyst called Claw "an important and logical next step" in a quote Oracle chose to publish in its own release. That is a vendor-selected quote, not an independent score.

What to do if you sell into these accounts

1. Match your product's core job against the 75 named apps.
2. Map what data your product needs and what it writes back.
3. Build a cross-stack story for buyers with agents in several systems.
4. Try the pro-code route yourself with Claude Code before a prospect asks. The which Claude model guide helps you pick one.
5. Ask Oracle for Claw pricing in writing.

If you are choosing among agent platforms for your own team, Smart Match turns a short questionnaire into a shortlist fitted to your stack. The best agentic AI roundup covers the wider field, and the real-money autonomy test shows how far unsupervised agents actually get. For a different take on delegated work, see Manus versus ChatGPT agent.

What to watch

Oracle AI World returns to Las Vegas from 25 to 28 October 2026, according to Oracle's newsroom, which makes it the likeliest place for Claw pricing, new models and customer numbers. Watch three things there:

• A price or a bundling statement for the Claw apps.
• The first named customer with a measured result.
• Any model beyond Gemini and OpenAI.

On the analyst side, a new Futurum edition would be the first independent read on whether four releases moved the needle.

The comparison set keeps moving too. Microsoft Copilot Studio, Microsoft Copilot and ChatGPT keep shipping agent features, and the Microsoft Copilot guide tracks where that lands for office workers. The directory pages for AI agents and agent records show the wider field.

The best single figure for a roadmap is still Oracle's claim of more than 1,000 agents running in live customer accounts as of its July release. An installed base is what a competing vendor has to route around.

What changed since 4 August

Re-checked on 23 October 2026 against Oracle's newsroom and the Futurum page.

• Added: the Claw runtime and its 25 apps, from 29 September.
• Replaced: the earlier line that Oracle had published no combined application total. Oracle now states 75 apps.
• Unchanged: Oracle and UiPath in Futurum's Aspiring zone, three vendors in Elite.
• Carried over from July and not re-verified: the certified-builder count above 80,000 and the 1,000-plus live agents, both Oracle's figures.
• Removed: the claim that Futurum updates this scorecard twice a year, which we could not source.

Frequently asked questions

What is Oracle's Agentic Applications Builder?

It is a feature of Oracle AI Agent Studio inside Oracle Fusion Cloud Applications for building AI agents natively. Oracle launched it as a no-code tool on March 24, 2026, then added pro-code development, including Claude Code and OpenAI Codex, on July 14, 2026.

What is Oracle Fusion Claw?

Fusion Claw is a governed agentic execution runtime that Oracle announced on September 29, 2026. A frontier model plans each job and deterministic computation executes it. Oracle launched 25 Claw-powered applications on it, running on Oracle Cloud Infrastructure with Gemini and OpenAI models.

Does Oracle charge extra for the Agentic Applications Builder or the Claw apps?

Oracle's March and July 2026 announcements say the builder carries no additional cost for existing Fusion subscribers. The September 29 Fusion Claw release states no price, and HokAI could not confirm that the new apps follow the same rule.

Is Oracle rated as a leader in agentic AI platforms?

Not by Futurum Group's Agentic AI Platforms for Enterprise Signal report, published June 1, 2026. It places Oracle in Aspiring, its lowest of four zones, beside UiPath, while Microsoft, Salesforce and ServiceNow sit in Elite. The report predates Fusion Claw.

How many Fusion Agentic Applications has Oracle released?

Oracle's September 29, 2026 release says its portfolio stands at 75 Agentic Applications, including 25 new Claw-powered ones. Earlier waves on April 9 added 12 apps for finance and supply chain, 8 for HR and 5 for customer experience.

Covered in this guide

• Claude Code: Anthropic's agent for the terminal, IDEs, desktop, web and mobile that edits files, runs tests and opens pull requests. There is no free plan.
• AWS: Amazon, founded in 1994, runs AWS, the cloud and retail company behind Bedrock, in-house Nova models and custom AI silicon.
• ChatGPT: ChatGPT is OpenAI's AI assistant, with 1.2 billion weekly users by OpenAI's count, the GPT-6 model family and plans that run from a free tier to a premium Pro tier.
• CrewAI: Open-source Python framework for multi-agent AI, MIT-licensed with about 59.6K GitHub stars, plus the CrewAI AMP platform for building and governing agents.
• Gemini 3.5 Pro: Google named Gemini 3.5 Pro at its May 2026 developer conference, but the model has not been released. Google has published no context window, price or benchmarks for it.
• Glean: Enterprise AI search platform at $300M ARR in 2026, connecting 100+ apps with permissions-aware results, AI agents, and MCP support from ~$50/user/month.
• Lindy: AI teammate that handles email, meetings and workflows across 1,000+ apps; no free plan, just a 7-day trial then per-seat billing.
• Make: No-code visual automation platform to connect apps and build AI workflows at scale
• Microsoft: Microsoft, founded 1975, is a public company (NASDAQ: MSFT) and the Redmond maker of Windows, Azure, GitHub and the Copilot assistants, with $331.8B of fiscal 2026 revenue.
• Microsoft Copilot: Your everyday AI companion powered by GPT-5.6 (preferred model July 2026), GPT-4o, and GPT-5 for productivity, creativity, and intelligent assistance
• Microsoft Copilot Studio: Build custom AI agents, workflows and apps without code, with MCP servers and credit-based billing billed through Azure.
• OpenAI Codex: Clones a GitHub repo into a sandbox, writes and tests code, then opens a pull request; OpenAI reported 20 million active users in August 2026.

Sources