AfterAgile™
Lens
Get the PDF

When AI Gets It Wrong

Correct the source, not just the output

AI will get things wrong – a requirement misheard in a session, a component invented, a persona that doesn't ring true. The rule is simple. Correct the source, not just the output, so it doesn't happen again.

The Correction Loop

  1. 1SpotThe owner vets it, or a test fails
  2. 2FlagMarked on the artifact, visible to all
  3. 3CorrectThe owner fixes the output
  4. 4TeachFix the source it learned from
  5. 5ConfirmRegenerate, and check it holds

Where Each Lesson Goes

When AI gets this wrongWho corrects itWhere the lesson goes
IntentCX and ProductThe context files, so the next one is generated right
The Collab BriefCXThe context files, and a note on how the session should be read
CodeEngineering, with the DefinerA new test, plus the guardrail or context file it broke
Designs and UIUXThe design system AI project
Persona walkthroughsUX, approved by CXThe personas themselves

Never Silent

Every correction is logged, so the team can see where AI is weakest.

Never Twice

If the same mistake comes back, the lesson didn't land in the right source.

Reviewed Thursdays

Corrections from the week are part of hour 2 of the Collab, Thursday afternoon.