Skip to content

AI Agent Readiness Check

Most agent projects stall as pilots, demos and internal tools nobody trusts enough to ship. The gap is rarely the model. It is whether the task is narrow, the sources are written down and current, and someone can show why the agent said what it said. Answer about the task you actually want to automate, not about AI in general.

1. Is the task bounded: one kind of output, one audience, a clear 'done'?
2. Does the knowledge the agent needs exist as documents or records, not in people's heads or chat threads?
3. Is that material current, with an owner and a date on each document?
4. Can every answer the agent gives be traced to a specific document and page?
5. When the answer is not in the sources, does the agent say so instead of guessing?
6. Is there a named person who signs off on the agent's output before it reaches a customer, auditor or regulator?
7. Can you show, for any past answer, why it said what it said: the question, the sources used, the version of the agent?
8. Do you know which data the agent must never see, and is that enforced by access rules rather than by instruction?
9. Do users know they are talking to an AI, and is that stated where they see it?
10. Is there a measured baseline for the task today, so you can tell whether the agent helped?
Readiness score

Gaps, with what ready looks like

    Where these questions come from

    They are the pattern behind the agents that made it past the pilot stage in our own work and in the research we track: task-specific scope, grounded answers with citations, an explicit refusal path, a human sign-off, and a log that an auditor or regulator can read. The disclosure question follows the EU AI Act transparency rules that have applied since 2 August 2026.

    Need grounded, cited answers from your own documents?

    See FanMind