From Tools Release to agentic AI
EnterpriseOne modernisation does not have to be one giant programme. Treat each layer as a separate upgrade with its own business case, technical boundary and proof of value.
Stage 1 — stabilise the traditional upgrade
Move the application and Tools baseline forward, migrate custom Objects deliberately, capture Version configuration and eliminate obsolete technical dependencies. This is where Boomerang, the Commanders, queue tools and attachment conversion belong.
Stage 2 — modernise identity
Give users a modern sign-on path and establish how enterprise identity maps to JDE and other protected resources. This becomes the trust foundation for later automation.
Stage 3 — make business actions explicit
Use Orchestrations for bounded business capabilities. Name them well, describe them well and give them contracts that applications and agents can understand.
Stage 4 — expose deliberate tool surfaces
Add MCP where runtime discovery and structured tool calling are useful. Keep servers focused: JDE execution, metadata, logs and files should not become one giant undifferentiated tool server.
Stage 5 — add the agentic layer
Introduce a conversational interface that can combine retrieval, reasoning and approved tool calls. At this stage, the AI is not being trusted with raw ERP access; it is choosing among capabilities you have already made explicit.
Stage 6 — automate recurring work
Once a workflow is predictable enough, scheduled or event-driven agents can execute it with the same policy and tool boundaries. The distinction between interactive and automated execution should remain visible in audit and governance.
A site can stop after identity, Orchestration or MCP and still have completed a useful capability upgrade.
The roadmap at a glance
| Stage | Primary question | Typical components |
|---|---|---|
| Release / Tools | Can we move forward with less retrofit risk? | Boomerang, Commanders, cleanup/conversion tools |
| Identity | Who is the user? | XpreSSO, ESI JDE SSO |
| Orchestration | What business actions are approved? | JDE Orchestrator, Composer |
| MCP | How does an AI client discover and call them? | Focused MCP servers, Steward |
| Agentic AI | How does a user ask, inspect and act? | BrainStorm or compatible assistant |
| Automation | Which recurring actions can run under policy? | Scheduled/event-driven agent workflows |