Upgrade JDE to agentic AI without replacing JDE.
The useful target is not an AI demo beside EnterpriseOne. It is an operating path where a user can ask a business question, inspect the evidence and invoke an approved JDE action while JDE remains the system of record and the security boundary remains explicit.
Start with a controlled action surface
Agentic AI becomes useful when it can do more than generate text. In an EnterpriseOne estate, the cleanest action boundary is usually a named JDE Orchestration: it has a defined purpose, known inputs and outputs, normal JDE execution semantics and an existing operational context.
That is a much better unit of AI action than unrestricted SQL, generic screen automation or a wildcard integration account.
Let the AI decide when to call an approved capability; let EnterpriseOne decide what that capability actually does.
The practical stack
| Layer | Role | Example component |
|---|---|---|
| User experience | Conversation, retrieval, citations and tool selection. | BrainStorm |
| JDE action design | Design and publish reusable business Compositions/Orchestrations. | Composer |
| Discovery | Expose JDE metadata and approved technical knowledge to the agent. | JDE Metadata MCP |
| Secure MCP gateway | Authenticate, govern and execute compatible MCP backends. | Steward |
| Identity | Modern sign-on and user identity into the JDE estate. | XpreSSO / ESI JDE SSO |
Do not begin with “give the model access to JDE”
That framing is too broad. The better sequence is: identify business questions, identify the approved actions needed to answer or complete them, publish those actions with narrow contracts, then expose them to the assistant through a governed tool interface.
This keeps the agentic layer understandable. It also means the same Orchestrations can be used by conventional applications, spreadsheets, automation and AI rather than becoming one-off AI-only integrations.
Where BrainStorm fits
BrainStorm provides the conversational layer, retrieval and tool orchestration. It can work with JDE identity, enterprise identity providers, Active Directory/Kerberos and remote MCP endpoints. The important point for an upgrade team is that this is an additive layer: the JDE application does not have to be replaced or rewritten to gain a modern AI interface.
Where Steward fits
Steward provides the secure MCP boundary for compatible tool servers and can preserve actual user authority when the downstream resource is Windows/AD protected. That matters when an agent needs to reach file shares, integrated-auth databases or other resources whose permissions already reflect the human user.