Content
Oracle AI World 2026 made it easier to modernize an Oracle estate in stages. You can now put AI on top of the systems you already run, with database-level controls, shared business definitions and joint EBS-Fusion analytics, instead of replacing those systems first. Much of the governance layer is still roadmap, though, so build your plan around what is GA today and what it costs to run.
Oracle held the event October 25-28 at The Venetian in Las Vegas. Our preview of Oracle AI World 2026 promised four checks for this recap: GA status, identity across agents, write support for open data, and customer evidence on upgrade cost and risk. You’ll find the answers below.
We judged each announcement by one test: does it lower the cost, risk or integration effort of running Oracle Database, EBS or Fusion?
The Headline from Oracle AI World 2026: AI That Acts Next to Your Systems of Record
Most CIOs have AI pilots that answer questions but can’t touch a transaction. Oracle’s pitch this year was AI that does the work inside the process, with the safety rules moved down into the database.
Three announcements carry that pitch:
| Layer | What Oracle showed | What changes for your estate |
|---|---|---|
| Execution | Fusion Companion coordinates transactions, workflows, approvals and agents. Oracle counts 25 Claw-powered apps as available. | AI moves from a side chat window into the workflow |
| Context | Oracle AI Data Platform holds business definitions that every agent reuses | One definition of revenue for every agent |
| Control | Deep Trust AI Architecture enforces access rules and business constraints in the database | A new AI access path hits the same rules as your apps |
This turns AI modernization into an ownership question. You decide which calls a model can make on its own and which permissions, approvals and transaction rules stay hard-coded.
What wasn’t new
Oracle AI Database 26ai and Autonomous AI Lakehouse debuted at AI World 2025, and Deep Data Security was first shown in April 2026. This year Oracle extended those pieces and wired them together.
The Announcements That Matter (and What You Can Use Today)
Roadmap features that slip into a project plan are a common reason budgets overrun. Only a handful of these announcements are deployable this quarter. Most semantic and governance features carry Oracle’s own “within the next 12 months” window, which is an estimate and not a contract.
| Announcement | Status (Nov 2026) | What it’s for |
|---|---|---|
| E-Business Suite Data Intelligence | GA | Shared analytics while you move from EBS to Fusion |
| FDI Data Agents | GA | Questions over Fusion data |
| Fusion Claw-powered apps (25) | Available, per Oracle | Specific ERP, HCM, SCM and CX tasks |
| Fusion Companion / Agentic Home | Part of release 26D | One interface over Fusion workflows |
| FDI Operational Analytics | Limited availability | Operational reporting |
| Deep Trust AI Architecture | Blueprint; check each component | Database-level control of AI access |
| Lakehouse Business Semantic Layer, FDI Deep Insight Agents | Planned, within 12 months | Shared KPIs, deeper Fusion analysis |
| AI Data Platform additions | Within 12 months; Governance and Control planned | Reusable context, agent oversight |
| Expanded Iceberg support | Announced, no date | Creating and changing Iceberg tables in SQL |
| Snowflake on OCI | Planned, no GA date | Snowflake next to your Oracle data |
| OpenAI: AI Workspace, App Composer, AI Security Mesh | Announced, no full GA schedule or price list | ChatGPT access, generated apps, security ops |
Deep Trust AI Architecture
Agents can’t skip rules that live in the database. Oracle’s blueprint stacks four layers. Autonomous AI Database patches the whole stack, but only on that managed service. Deep Data Security checks the identity token from your directory, and SQL Assertions block changes that break business constraints. Deep Trust Data APIs give agents business objects instead of raw tables. At the top, AI writes APEXlang that the APEX runtime executes.
Your team still writes the policies and makes sure the right identity reaches the database.
Autonomous AI Lakehouse and Oracle AI Data Platform
Business definitions become a shared asset that both BI and agents draw on. The Lakehouse Business Semantic Layer stores dimensions, measures and KPIs. Expanded Iceberg support lets you create and change tables in SQL. That settles one preview question: write support is growing, with no date attached.
Oracle AI Data Platform drafts business ontologies with AI from your schemas, semantic models and API docs, and your domain experts sign them off. Its AI Registry lists agents, tools, MCP servers, owners and permissions. The registry records who has access. Enforcing it falls to Governance and Control, which is still planned.
Fusion agents and implementation automation
Oracle is going after the cost of Fusion projects. The Fusion Setup and Configuration Registry turns setup knowledge into executable skills in a public Git repository, and AI Success Navigator runs design, configuration and testing agents on it. Oracle’s customer example, Lexington Health, saved 80% of the effort on certain design and configuration tasks. Read that as a task-level number. It says nothing about total project cost.
EBS and Fusion coexistence
For EBS customers, this is the most usable news of the week. E-Business Suite Data Intelligence is GA. Managed pipelines and prebuilt analytics span EBS and Fusion during the migration, so finance gets one set of reports long before cutover. Confirm it covers your EBS version and customizations.
Snowflake on OCI and the Oracle OpenAI deal
Both widen your options. Your integration work stays.
Snowflake on OCI is a planned first-party service that you buy and pay for through Oracle, Universal Credits included, with integrations to Fusion, NetSuite and Oracle databases on the roadmap. InfoWorld lists the trade-offs: data still gets copied, and you run two catalogs on two bills. If that move is already on your table, our Oracle to Snowflake migration roadmap walks through it.
The Oracle OpenAI collaboration brings ChatGPT-based AI Workspace, App Composer for Codex-built apps on OCI, and AI Security Mesh, whose agents get scoped identities and need your approval for each fix. Before anyone signs, find out what data leaves Oracle and which contract governs it.
Enterprise Impact: Enterprise AI Platform and Data Governance
Buying an AI platform doesn’t make AI trustworthy for your auditors. Oracle’s enterprise AI platform is a stack of layers, and each one hands you something new to run.
| Layer | Oracle provides | You still own |
|---|---|---|
| Business context | Ontologies, metrics, knowledge graphs | Definitions, sign-off, change control |
| Data access | Database policies, business-object APIs | Least-privilege design, supported integrations |
| Agent execution | Fusion workflows and agents | Agent limits, exception handling |
| AI inventory | AI Registry | Registering every agent and naming an owner |
| Evaluation and ops | Agent tests, errors, latency, AI Units | Test sets, release gates, chargeback |
You get reuse. In return, semantics, policies and test sets need owners and budgets, the same test we apply to any enterprise AI platform.
Data governance covers more than access
AI agent governance has to settle five things:
- Authorization: what this user or agent can reach
- Integrity: whether a change breaks a business rule
- Meaning: which definition of revenue or customer applies
- Quality: whether answers pass reference tests
- Accountability: who owns the agent and its permissions
They fail independently. A query can be fully authorized and still return the wrong number. A correct number can still trigger an action no one should take. Our guide to data governance for AI covers the setup.
Identity: the preview question, answered in part
Inside the database, identity holds. Through external assistants, it’s still unclear. Deep Data Security verifies the end user’s directory token, and Security Mesh agents run under their own scoped identities. Oracle hasn’t fully documented how the user’s identity travels through ChatGPT-based AI Workspace or third-party MCP tools.
NL2SQL is a knowledge-engineering job
Deep NL2SQL is only as accurate as its business knowledge graph and its regression tests against approved “gold” SQL. Budget for curating terms and expected answers, not just for the model.
Gaps, Trade-Offs and Open Questions
Availability, cost and proof are where this year’s story is thinnest. Commit to what is GA and carry the rest as roadmap risk.
Upgrades still carry risk
We found no independent customer data on upgrade cost or downtime at the event. The documentation sets three facts:
- You can upgrade to 26ai directly from 19c or 21c. Older releases need an intermediate step.
- Raising the
COMPATIBLEparameter removes downgrade and flashback to restore points. Test first and raise it last. - EBS 12.2 is certified on 26ai for on-premises Linux x86-64, with listed conditions. Other platforms and integrations need their own check.
Openness and governance stop at the boundary
Iceberg makes your data format portable. Metadata, definitions, policies, agent workflows and runtimes each need their own exit plan. A policy enforced in Oracle Database also won’t follow a copy into Snowflake or a prompt into ChatGPT, so test policy mapping and identity at every handoff.
Cost runs on several meters
Oracle bills AI Data Platform consumption apart from storage, logging, generative models and databases. The default Master Catalog has a floor of 230 AIDP Units per hour, about $170 a month at the list price of $0.001 per unit, before any workload runs. Fusion uses a separate meter, AI Units, based on tokens. Every saving Oracle points to, from less configuration to fewer data copies, has to be netted against validation work, connector gaps and the people who maintain definitions.
Questions to put to Oracle
- Which release, region and deployment model supports each component today?
- What is included, licensed separately or metered?
- How does user identity pass through ChatGPT, MCP tools and database connections?
- Which write paths are supported for EBS, PeopleSoft and JD Edwards?
Practical Next Steps for Your Oracle Estate
Pick one estate profile and one bounded workflow before you look at an agent catalog.
Pick your modernization path
Your hybrid cloud strategy depends on where you start:
| Your estate | First move | Check first |
|---|---|---|
| Fusion customer | Extend selected workflows with Companion and Claw apps | 26D entitlement, approvals, evaluation |
| EBS moving to Fusion | Add EBS Data Intelligence during coexistence | Version coverage, historical data, reconciliation |
| Self-managed Oracle Database | Upgrade to 26ai, then expose business-object APIs where needed | App certification, COMPATIBLE timing, licensing |
| Oracle on OCI, Snowflake elsewhere | Evaluate Snowflake on OCI once it’s GA | Regions, feature parity, transfer costs |
| AWS- or Azure-centered | Compare existing multicloud Oracle options before adding OCI services | End-to-end architecture |
An Oracle EBS to cloud migration and an Oracle database migration are separate decisions with separate budgets.
Run a 90-day plan
- Days 1-15: Inventory releases, apps, integrations, data copies and service accounts. Choose one business problem and log the status and price of every component it relies on.
- Days 16-30: Set permissions, business definitions, gold-answer tests, approval limits and a cost baseline.
- Days 31-60: Pilot read-only or in non-production. Feed it ambiguous requests, stale data, unauthorized requests and failures.
- Days 61-90: Compare against the baseline and scale only where cost, quality and control targets hold.
Finance questions checked against approved SQL make a good first pilot. Let agents read, then recommend, then act with approval, and track cost per correctly completed task and wrong-answer rate. Our AI-ready data checklist covers the groundwork.
Key Takeaways from Oracle AI World 2026
- You can modernize in stages and add AI without replacing core systems first.
- Rules sit in the database, so they apply to every agent. Writing them is still your job.
- Business definitions need owners, because semantic layers are now production assets.
- Integration gets easier and stays hard. Copies, identity gaps and second catalogs remain.
- Price the full stack. Much of the governance layer is still planned.
Which Oracle announcements belong in your plan?
We map your Oracle databases, EBS or Fusion apps and integrations, sort what you can deploy now from what is roadmap, and give you a costed path to upgrade, coexist or move.
Starts with a 30-minute call about your estate.
Starts with a 30-minute call about your estate.
FAQ
What did Oracle announce at Oracle AI World 2026?
The main items were Deep Trust AI Architecture, a Lakehouse semantic layer, new Oracle AI Data Platform features, Fusion agents, E-Business Suite Data Intelligence, a planned Snowflake service on OCI and three OpenAI products, including AI Workspace. Many are scheduled within 12 months.
Is Snowflake available on Oracle OCI?
Not yet. Oracle and Snowflake announced a planned native service on OCI, sold and billed by Oracle. No GA date had been confirmed as of November 2026.
Can you upgrade Oracle 19c directly to 26ai?
Yes. Oracle supports direct upgrades to 26ai from 19c and 21c, while older releases need an intermediate upgrade. Raise the COMPATIBLE parameter only after testing, because it blocks downgrade.
Do EBS customers need to move to Fusion to use Oracle AI?
No. E-Business Suite Data Intelligence is GA and covers analytics across EBS and Fusion during a phased move, and EBS 12.2 is certified on Oracle AI Database 26ai with conditions. Confirm supported APIs before letting agents write back into EBS.
Sources
- Oracle AI Data Platform Brings AI-Assisted Business Context to Enterprise AI
- Oracle Advances Autonomous AI Lakehouse with Business Semantic Layer
- Oracle Expands Fusion Data Intelligence
- Oracle Unveils Agentic AI Platform to Accelerate Fusion Implementation
- Announcing Oracle Deep Trust AI Architecture
- Oracle Reinvents Enterprise Computing for the AI Era with OpenAI
- Oracle Advances Enterprise Security for the AI Era
- Oracle Expands Native AI and Data Services on OCI with Snowflake
- Snowflake: Snowflake and Oracle Partnership
- InfoWorld: Oracle pushes AI deeper into enterprise workflows with Fusion Companion
- InfoWorld: Oracle adds native Snowflake service to OCI
- Oracle: Releases That Support Direct Upgrade (26ai)
- Oracle: What Is Database Compatibility
- EBS 12.2 Certified with Oracle AI Database 26ai
- Oracle AI Data Platform Pricing
- Oracle Cloud Price List
- Monitor and Evaluate AI Agents – Oracle Fusion AI
