Skip to content
← Research

Decision framework · Illustrative

Where Should an AI Agent Sit in Your Technology Stack?

A practical way to connect AI to existing systems without losing control of the work.

The short answer

Start with a bounded task and the systems that hold its evidence. Give the AI only the access it needs, define where human approval is required, and measure whether the work actually improves.

See what happens

A bounded task
Approved data and tools
AI proposes a next step
Human approval where required
Action and audit record

A conceptual architecture. An AI agent should not receive broad access merely because the task spans several systems.

Systems involved

Where it breaks

Design choiceFailure to look for
Unclear task boundaryThe agent keeps acting without a useful stopping point.
Unreliable source dataA convincing answer is built on the wrong record.
Broad write permissionsA mistake can change stock, orders or customer commitments.
No evaluationActivity is mistaken for operational value.

How companies approach it

  • Choose one workflow with a visible manual baseline.
  • Begin with read-only investigation or drafting.
  • Define allowed tools, escalation and approval points.
  • Log the evidence, tool calls and outcome.
  • Measure time saved alongside error rates and review effort.