What if AI can do more than answer?
EKA primarily introduces the idea of AI reading, retrieving and reasoning over the enterprise.
A later step is an enterprise agent that may also participate in business work.
That changes the architecture substantially.
Read/reason vs act
- READ
- RETRIEVE
- REASON
- ANSWER
- OBSERVE
- REASON
- DECIDE / PROPOSE
- ACT
The important new word is:
ACT
because actions can change enterprise state.
The rule that should not change
A future agent should not gain arbitrary database write access.
It should use the same controlled business mutation path as other callers.
The caller may be intelligent.
The enterprise rules still belong to the handler.
Why this matters
Suppose an agent decides:
The unsafe approach would be:
agent → directly edit ERP tables
A safer enterprise architecture is:
Use the existing mutation architecture
The Day-0 Explorer does not currently expose an agent-action record.
So the Guide should not fabricate one.
Instead, show the already established command/handler path and explain:
Future agents would enter through this same governed business boundary.
Agent does not mean unrestricted autonomy
Enterprise agents may eventually operate at different levels.
For example:
The Guide should not claim one final autonomy model until agent architecture is deliberately designed.
Business time still matters
If an agent acts inside a simulated runtime, the action still occurs in enterprise business time.
Its consequences may still:
- create source changes
- trigger events
- become relay work
- be extracted later
- reach canonical later
An agent does not bypass the runtime model.
Provenance becomes even more important
When AI can change state, the enterprise should be able to answer:
- Which agent acted?
- What did it observe?
- Which action did it invoke?
- Under whose authority?
- What command executed?
- What event resulted?
- What changed downstream?
The existing command/audit/event architecture is the foundation for that future capability.
A future action loop
A future action loop
This is a future architecture direction.
It is not current Day-0 functionality.
Chapter takeaway
- EKA reads and reasons.
- Future enterprise agents may also act.
- Actions should still pass through commands and handlers.
- AI should not bypass authorization, validation, idempotency or audit.
- The agent becomes another governed participant in the enterprise.