AI Is Mostly an Integration Problem
Many enterprise AI conversations begin with the model. Which model should be used? Which provider is best? Which assistant or agent platform should be selected?
Those are useful questions, but they are rarely the whole problem.
The visible feature may be AI. The delivery challenge is usually everything around it.
In real organisations, AI has to authenticate, retrieve context, respect permissions, interact with workflows, create an audit trail, handle uncertainty and fail safely.
The hidden platform
Successful enterprise AI depends on a hidden platform of identity, integration, data access, observability, operational ownership and human escalation.
The model may interpret, summarise or recommend. The enterprise architecture must decide how that output enters the operating model.
Where AI projects often struggle
- Fragmented systems and inconsistent APIs.
- Unclear ownership for AI-assisted outcomes.
- Weak auditability from prompt to downstream action.
- Uncontrolled vendor integrations into core platforms.
- Support teams unable to trace what happened and why.
Architecture implication
The architect should not treat AI as a bolt-on feature. It needs to sit inside a governed ecosystem that can be monitored, supported, evolved and challenged.
AI often becomes the visible capability. Integration remains the hidden platform.