Oracle Fusion 26D Readiness Checklist: Treat AI Agents as Decisions, Not Features

To prepare for Oracle Fusion 26D, confirm your cohort dates, then run every AI agent through a five-question test (entitlement, access, boundaries, control, ownership and cost) and give each one outcome: enable now, pilot, defer or hold.

Key takeaways

  • Oracle’s 26D Financials update lists 17 AI agent features, and none arrives switched on. Each needs setup, a service request or an opt-in (Oracle 26D Financials Feature Summary, 2026).
  • Test environments update on the first Friday of your cohort month. Production follows two weeks later, on the third Friday.
  • Switching on one agent typically takes five workstreams with different owners. Without a coordinator, agents stall or go live without controls.
  • Role-based security is the real control. An agent is only as safe as the roles of the people using it.
  • Every agent should get a named owner, a decision (enable, pilot, defer or hold) and, for custom agents, an AI unit budget.

What is Oracle Fusion 26D?

Oracle Fusion 26D is the fourth 2026 quarterly update to Oracle Fusion Cloud Applications. It reaches customer environments by cohort between November 2026 and January 2027. Oracle names each update by year and order, so 26A was the first update of 2026 (Oracle Fusion Insider blog, 2026).

Quarterly readiness used to mean reading the What’s New notes and regression testing the screens that changed. That is no longer enough. A growing share of each update is AI agents that act on your data, and each one raises questions a regression test cannot answer: who may use it, what can it touch, and who is accountable for what it does?

Oracle’s 26D Financials feature summary lists 17 AI agent features across cash processing, payables, payments, assets, budgetary control, expenses, joint ventures and treasury. Fifteen need setup, one needs a service request and one is an opt-in (Oracle 26D Financials Feature Summary, 2026). Nothing is on by default, so every agent is a decision someone must own.

When does Oracle Fusion 26D reach my environments?

Oracle updates test environments on the first Friday of your cohort month and production two weeks later, on the third Friday. Cohort A updates in November, Cohort B in December and Cohort C in January 2027 (Oracle Fusion Insider; ERP Today, Aug 2026).

Oracle Fusion 26D update dates by cohort
Cohort Test environment Production
A Fri, Nov 6, 2026 Fri, Nov 20, 2026
B Fri, Dec 4, 2026 Fri, Dec 18, 2026
C January 2027 January 2027

Middle East regions update on the Thursday before. Confirm your cohort and exact dates in the Oracle Cloud Console.

Treat the two weeks between test and production as a managed stage, with owners, scope and a change freeze agreed in advance.

What does it take to switch on one Oracle AI agent?

Switching on one agent usually takes five pieces of work owned by different people: an environment check, configuration, security, AI administration and data readiness. “Setup Required” in the release notes understates this.

The pattern below comes from the HCM agents Oracle documented in its 26B and 26C updates. Each agent has its own setup page, so check yours.

Five workstreams behind one Oracle AI agent
Workstream Typical owner What Oracle’s documentation asks for
Environment check IT or Oracle Support liaison Your help desk asks Oracle Support to verify the environment before setup
Configuration Functional or technical admin A Security Console integration profile option is set to Yes
Security Security and role owner Permission groups are enabled for the right roles, with agent access duties added where needed
AI administration Oracle AI Agent Studio administrator Each role is explicitly granted access, and you create your own agent from the preconfigured template
Data readiness Process owner Some agents depend on data refresh processes before they show current results

Agent readiness has no single module owner. If nobody coordinates these workstreams, nothing gets switched on, or something gets switched on without the right controls.

How do you decide whether to enable an Oracle AI agent?

Before switching an agent on, answer five questions: are we entitled to it, who can access it, what are its boundaries, are controls in place, and who owns it and its cost? This is dotSolved’s minimum test.

dotSolved’s five-question agent test
Question What to confirm
Entitlement Are we licensed to use it? Check agent and agentic application entitlements with Oracle before building a business case.
Access Which roles can use it, and what can those roles see and change?
Boundaries What can the agent not do? Read the documented limits before promising value to the business.
Control Are approvals, guardrails and test evidence in place?
Ownership and cost Who is accountable for it, and what AI usage allowance applies?

What are the four outcomes for each agent?

Every agent should leave the five-question test with exactly one outcome: enable now, pilot, defer or hold. “We’ll look at it later” is not an outcome.

The four outcomes for every Oracle AI agent
Outcome When it applies
Enable now Low data risk, a clear owner, tested with real roles
Pilot It writes to transactions or needs approvals. Start with a small group and review results.
Defer Entitlement or integration impact is unclear. Revisit next quarter.
Hold No named owner, or the access model is not clean

Why is role access the real control for Oracle AI agents?

Oracle’s agents work inside your existing data security, so users see only what their roles allow. That makes role design, not the agent, the control that matters most.

Oracle’s Person Data Assistant shows how this works. It applies role-based data security, and an administrator must grant each role access to the agent explicitly. The seeded HR Specialist role already carries the required duty. Employee, line manager and custom roles need it added, and line managers also need a worker search privilege (Oracle HCM documentation, 26B–26C).

That is good design, with a consequence. An over-privileged role was always a risk. With agents, it becomes a faster route to a mistake. Review the roles of your first pilot users before the test update, and test with real roles, not administrator accounts.

What can’t an Oracle AI agent do?

Every Oracle agent has documented limits, and business users should hear them from you before they meet them in production. Read the boundaries on each agent’s page before you promise value.

Oracle’s Line Manager Salary Change agent is a good example. As documented at release, it:

  • works on a manager’s direct reports, one employee at a time, and one salary field per request;
  • follows the standard approval workflow and checks for pending salary transactions;
  • rejects past effective dates and does not convert currencies;
  • returns a generic error when a submission fails.

These are sensible controls. They also mean the agent will not do everything a manager might expect.

How do you test custom agents in Oracle AI Agent Studio and control cost?

Regression test every agent you build with Oracle’s ATLAS testing suite, and agree an AI unit allowance per workflow before go-live. A test that passes can still hide an agent that has quietly become expensive.

If your team builds or tailors agents in Oracle AI Agent Studio, 26D adds controls worth using: guardrails, model lifecycle management, AI unit budgets and release pipeline integration. ATLAS (Agentic Testing and Lifecycle Automation Suite) replays recorded scenarios, checks routing and response quality, compares models and reports AI unit usage (Oracle Fusion CoE blog, 2026). Oracle meters model usage in AI units, so cost belongs in the test plan, not just the invoice.

What is being retired in Oracle Fusion 26D?

Read the Important Actions and Considerations page for every product you run, because retirements and deprecations land there. Two examples to check:

  • Personal Information Assistant template. Oracle has said it will no longer be enhanced and recommends the Person Data Assistant for new builds. Existing agents keep working.
  • BI Cloud Connector view objects. Oracle publishes a list of deprecated and new BI view objects. Review any extracts that depend on them.

Oracle Fusion 26D readiness checklist: 10 steps

  1. Confirm your cohort and update dates in the Oracle Cloud Console.
  2. Capture a baseline of roles, privileges, profile options and enabled offerings before the test update.
  3. Read each product’s 26D What’s New and Important Actions pages, and list every feature that needs an action.
  4. Name a business owner for every opt-in and every agent.
  5. Run the five-question test on each agent.
  6. Assign one outcome: enable now, pilot, defer or hold.
  7. Clean up the roles of your pilot users first.
  8. Test as a manager, a specialist and an analyst, not as an administrator.
  9. Regression test any agent you build, and set an AI usage allowance.
  10. After production, confirm configuration matches what you tested, and review agent results monthly.

Frequently asked questions

Are Oracle Fusion 26D AI agents turned on automatically?

No. In Oracle’s 26D Financials feature summary, every AI agent feature needs an action before use: setup, a service request or an opt-in. Nothing changes for users until someone in your organization enables it.

How long do I have between the 26D test and production updates?

Two weeks. Oracle updates test environments on the first Friday of the cohort month and production on the third Friday. Middle East regions update on the Thursday before each date.

Who should own Oracle AI agent readiness?

Name one business owner per agent and one coordinator across the five workstreams. Agent setup crosses IT, functional administration, security, AI Agent Studio administration and process ownership, so no single module owner covers it.

Do Oracle AI agents bypass role-based security?

No. Oracle’s agents work within existing data security, so users see only what their roles allow. That is why over-privileged roles become a bigger risk once agents are switched on.

What does it cost to run custom agents in Oracle AI Agent Studio?

Oracle meters model usage in AI units. Check your entitlement with Oracle, set an AI unit allowance per workflow, and use ATLAS reports to catch agents whose usage drifts upward.

How dotSolved can help

dotSolved’s Oracle practice runs a 26D agent decision workshop. We work through the five-question test for the agents in your modules and leave you with a decision register: what to enable, pilot, defer or hold, with owners and test scope. It is advisory-led and built for pilot to production.

Explore More Blogs

  • Enterprise AI Orchestration
    Turning AI into Business Operations

    Read More
  • The New AI Workforce
    Why Workforce Readiness Determines AI Success

    Read More
  • AI Development Lifecycle
    How to Build Enterprise AI That Compounds

    Read More