AI Agent Studio arrives free with every Fusion subscription, templates included. This guide opens the box and marks the four places cost appears anyway, so that a tool needing no purchase order does not quietly write your next invoice.
Oracle AI Agent Studio arrives with every Fusion Cloud subscription at no license cost, templates included. What sits outside that free boundary is the consumption your agents generate and the entitlement that governs custom agents in production. This guide opens the box, item by item, and marks where money starts.
It is the environment for configuring, extending and orchestrating AI agents inside Fusion Applications, and it arrives with the Fusion Cloud subscription you already hold at no separate license charge. Oracle set out the launch detail on the Oracle Fusion Insider blog.
Three ideas have to stay apart to answer the cost question honestly. The tooling and the templates are free, the work the agents do is metered, and running custom agents at production scale is separately entitled.
Consumption and the production entitlement. Every action an agent takes draws against the AI Unit meter once whatever monthly grant your subscription carries has been spent, and custom agents operating at production scale sit behind a separate subscription. Oracle marks that second boundary in its Custom AI Agent subscription documentation.
Neither of those is a trap. Both are simply invisible from inside a tool that opens without a purchase order, which is exactly why they surprise people.
A library of prebuilt patterns mapped onto standard Fusion processes, which is why most teams start by configuring rather than constructing. The library is the fastest route to something demonstrable and the cheapest thing to operate. It is also the part of the offer buyers consistently underuse.
Where the shipped templates land across the Fusion pillars
| Pillar | Typical shipped pattern | What it displaces | Where money appears |
|---|---|---|---|
| HCM | Employee question handling and onboarding assistance | Tier one service desk tickets | High volume, so tier choice dominates |
| ERP | Invoice narrative and approval summarization | Manual review commentary | Period end spikes drive the peak |
| SCM | Supplier and order status summaries | Status chasing by email | Scheduled runs create a floor |
| CX | Service case drafting and triage | First draft response writing | Grounding breadth drives the draw |
More than the business case usually assumes, and the gap is worth measuring before anybody writes code. In the rollouts we have reviewed, the first workshop produces a long list of agent ideas and a strong instinct that most of them are bespoke.
They rarely are. Score the list against the shipped library using three questions, and the bespoke column tends to shrink to a handful of genuinely distinctive processes.
Run that scoring in an afternoon with the process owners in the room. It is the cheapest hour in the entire program and it usually removes most of the custom pipeline.
Five parts, and only two of them are free to change without a cost consequence. Knowing which is which is the whole discipline.
Change the instructions and the presentation, leave the tools and the grounding alone, and never quietly promote the model tier while you are testing. Those three rules keep an extended template behaving like a shipped one.
The common failure is subtler than a big rewrite. Somebody adds one more lookup to make an answer richer, the agent now performs an extra retrieval on every run, and a small quality improvement becomes a permanent multiplier on a high frequency process.
At two moments, and opening the tool is neither of them. The first is when agent activity passes whatever monthly unit grant your subscription carries, and the second is when a custom agent moves into a live environment and crosses the separate entitlement line. Oracle's capability material sits on its AI Agents for Fusion Applications page.
Free tool, four cost drivers, ranked by how much they move
| Driver | What it is | How to see it early | The control |
|---|---|---|---|
| Model tier | Which model the agent reasons on | Run the same task on two tiers and compare draw | A written model policy, enforced at design review |
| Grounding breadth | How much data is read before answering | Count retrievals per run in the test pod | Scope grounding to the narrowest useful set |
| Agent chattiness | Retries, replanning and multi turn loops | Look at the spread, not the average, across runs | Cap steps per run and log the outliers |
| Promotion to production | Moving a custom agent into a live pod | Nobody sees it. It is a permission, not a purchase | Approval gate on promotion, named owner per agent |
Notice that three of the four are design decisions taken by people who have never seen an Oracle invoice. That is the reason a purely commercial control fails here, and a design review control works.
Source: Redress Compliance advisory engagement file covering 2024 to 2026, checked against Oracle product documentation.
Three questions, asked at design review, settle it before anybody has spent anything. Any agent answering yes to all three will stay unremarkable on your bill for as long as it lives.
An agent that answers no to any of the three is not necessarily wrong. It simply needs a named owner who can explain what the extra spend is buying, which is a conversation worth having once rather than a surprise worth having annually.
The advice you will hear from every direction is that the studio is free, so you should start building custom agents to differentiate your processes. We disagree, and not for compliance reasons. The template you configure and then leave alone is the cheapest asset in this whole stack, it survives quarterly updates with the least attention, and it stays comfortably on the free side of the entitlement line. Custom construction is where consumption climbs, where regression testing lands on your team every quarter, and where the separately subscribed boundary sits. Build custom where the process genuinely is your advantage. Everywhere else, the unmodified template is the sophisticated choice, not the lazy one.
Capability, inclusion and the custom agent boundary. Not commercial terms. Oracle's product and announcement material describes what the studio does and states that it comes with Fusion, and it stops there.
Nothing on a public Oracle page sets out a unit rate, a monthly allowance figure, or a pack price for the agent meter. The numbers circulating in the market come from price list extracts and third party analysis, and they are corroborated rather than published.
Four documents to read before anyone builds anything
| Document | What it settles | Does it bind Oracle? |
|---|---|---|
| Custom AI Agent subscription guidance | Where the free studio stops and entitlement begins | No. It is guidance, and it can be revised |
| Fusion Cloud global price list in force on your order date | The rate and pack shape at that moment | No. It is a snapshot, revised without notice |
| Your ordering document | What you are actually entitled to and at what price | Yes. This is the one that counts |
| The applicable cloud service descriptions | What the service is, and what Oracle may change | Yes, and it is the one nobody reads |
The practical instruction is short. Read the two that bind, treat the other two as weather reports, and do not let a business case rest on a figure your own paperwork does not repeat. The current Fusion Cloud global price list is the place to check the shape on the day you sign.
By governing the design, because there is no purchase to intercept. A free tool defeats every control your organization normally relies on: no requisition, no approval, no vendor conversation, no budget line. The first three agents will exist before procurement hears the word.
Write one page. Name the default tier for routine actions, the conditions that justify a higher tier, and the person who signs that exception. Then attach it to the design review checklist so it is applied by the people making the choice.
This single artefact does more for the eventual bill than any amount of negotiation. Rates are argued once a year. Tier defaults are exercised every hour of every day, by every agent your team has built.
Who owns which decision, and what happens when nobody does
| Decision | Right owner | What happens by default |
|---|---|---|
| Model tier per agent | Platform owner, against a written policy | The builder picks the tier that demos best |
| Grounding scope | Process owner with the data owner | Scope creeps outward one source at a time |
| Promotion to a live pod | Release manager, with a named approver | Whoever has the permission decides alone |
| Retiring unused agents | The agent owner, reviewed quarterly | Nothing is ever retired |
| Consumption reporting | Finance, using platform data | The first real number is an invoice |
They treat template based agents and custom agents very differently, and the difference lands on your calendar rather than Oracle's. Shipped patterns move forward with the release. Anything you constructed yourself is yours to revalidate.
It is the build layer. The unit meter is the charge on what you build, and the Agentic Applications platform fee is the gate above both that governs publishing composed applications to production. Three layers, three different commercial behaviors.
The license is free. The bill is written by the model your agents think on and the day one of them goes live.
Read the three together and the economics stop being mysterious. The Fusion AI Agents pillar prices the meter and sets out the full model, and the Agentic Applications subscription guide covers the platform fee and what triggers it.
Oracle's own account of how the layers expanded sits in the Agentic Applications Builder announcement of March 2026, which is the moment the build layer and the publishing layer were formally separated.
Run this sequence in the first month of any studio rollout. It is deliberately cheap, and every step produces evidence you will need at the next renewal.
Yes, in the sense that matters to a license audit: it arrives with a Fusion Cloud subscription and carries no separate license charge, templates included. What it does not include is the consumption your agents generate and the entitlement required for custom agents at production scale. Free tool, metered output.
The configuration environment, the shipped template library across HCM, ERP, SCM and CX, the orchestration tooling for chaining agents together, and model selection. Testing outside production is included too, which is where any sensible first phase belongs. Model selection is the item that decides your running cost.
Not to use the studio itself, but yes for custom agents running at production scale, which Oracle covers in its own guidance. A prototype built in a test pod sits inside the free boundary. The same prototype moved into a live pod is a different question, and the move is a permission rather than a purchase.
Four things: the model tier the agent reasons on, how much data it reads before answering, how many retries and replanning loops it performs, and whether a custom agent gets promoted into a live environment. Agent count barely matters by comparison. Tier and grounding breadth do most of the work.
Configure the templates wherever the process is not genuinely your competitive advantage. An unmodified template is cheaper to run, survives quarterly updates with less attention, and stays on the free side of the entitlement boundary. Reserve custom construction for the handful of processes where difference actually earns something.
The studio is the free build and orchestration layer; the Agentic Applications subscription is a distinct platform fee governing publication of composed applications into a live pod. The studio produces agents, the unit meter charges for their work, and the platform fee opens the Builder and the production publishing right. Three layers, three separate lines.
No public Oracle page states a unit rate, an allowance figure or a pack price. The figures circulating in the market are corroborated by more than one third party, but they come from price list extracts rather than Oracle publication. Use them for orientation, and require your own ordering document to restate whatever you rely on.
Shipped patterns move forward with the release, while anything you constructed yourself is yours to revalidate. Keep a note of the release each custom agent was validated against so regressions have a baseline. Budget the revalidation effort as a standing cost, because it recurs every quarter for as long as the agent lives.
The AI Unit cost model, the free allowance math, the LLM cost trigger, the Agentic Applications platform fee, and the 26C levers to set before you upgrade.
Used across more than five hundred enterprise engagements. Independent. Buyer side. Built for procurement leaders sizing Fusion AI.
500+ enterprise clients. 11 vendor practices. We will map which agents draw units, which need a custom subscription, and how to keep the studio close to free.
One email a month on Oracle AI Units, Fusion renewals, ULA certification, and audit defense. Buyer side only.
