There has recently been significant discourse about AI in PeopleSoft, and much of it overstates what has actually shipped. All else aside, this is the bottom line: PeopleSoft does not come with AI built in. What Oracle has delivered is the secure infrastructure that lets PeopleSoft connect to whichever AI models and agents your organization chooses or already utilizes.
That distinction matters. It changes what you should expect, what you need to decide, and what it takes to garner value from AI in PeopleSoft.
What's real
Oracle has been clear about its approach. Generative AI will not be embedded directly in PeopleSoft applications. Instead, PeopleTools 8.63, now available on OCI and on-premises, provides building blocks that connect PeopleSoft to the AI platforms and models each customer chooses, while respecting PeopleSoft security.
The confirmed building blocks:
MCP server support. AI agents can find and use PeopleSoft APIs. Administrators choose which APIs are exposed, and PeopleSoft security and business rules still apply.
Natural Language Search Service. Answers plain-language questions using PeopleSoft Search, through a REST API. Users see only results they already have access to.
MCP server for PeopleTools metadata. Lets AI tools read definitions such as records, pages and PeopleCode, which helps with code review, debugging and documentation.
Application Designer AI Assist. A chat window that explains, fixes and writes PeopleCode, using the customer's choice of LLM.
Magic Button. A button next to long text fields that runs an AI action, set up through configuration rather than page customization.
Three principles sit underneath all of this: no vendor lock-in, PeopleSoft as the secure system of record, and configuration over customization.
What's overstated
Most of the overstatement comes from treating described possibilities as delivered features. Four claims come up often.
"PeopleSoft now has AI built in." It does not. There is no AI model inside PeopleSoft. The value depends on the AI services you connect and the use cases you configure.
"AI agents can now run your finance processes." Agents can complete transactions through PeopleSoft APIs that administrators choose to expose. Oracle has said this could extend to expenses and vouchers. That is a possibility Oracle has described, not a delivered example.
"AI can see everything in PeopleSoft." The opposite is true by design. Search returns only what the signed-in user is permitted to see, and agents work only through the APIs an administrator exposes.
"AI will rewrite your custom code." AI Assist can explain, fix and write PeopleCode. Before release, Oracle said the first version would cover a limited set of objects and only save changes to PeopleCode. It is a development aid, not an automated rewrite.
What it means for your organization
AI in PeopleSoft is a set of decisions, not a feature you switch on. You choose the AI services and models. You decide which APIs agents can use. You pick the use cases worth configuring.
Oracle's own guidance on where to start is practical: repetitive tasks, high-volume questions, search pain points, development work, and processes suited to natural-language or agentic experiences. Realistic early examples include an assistant that answers questions using PeopleSoft data a user can already see, summarizing long text entries, and helping developers understand older custom code.
The opportunity is real. It is just narrower, more deliberate and more secure than the headlines suggest, and that is a good thing for organizations that rely on PeopleSoft as their system of record.
