Lessons
1Find the first wrong transition35 min read
Diagnose an agent incident from a minimal trace.
- →Locate the first incorrect transition in an agent trace
- →Distinguish tool failure from model decision failure
- →Report uncertainty and ownership in an incident answer
2Build an outcome verifier that cannot be fooled by narration35 min read
Define success, partial success, and invalid shortcuts for a task.
- →Design a verifier that measures the requested outcome
3Defend architecture with a failure budget35 min read
Choose a workflow or adaptive agent for a stated task distribution.
- →Choose an agent architecture from task constraints
4Make a release decision from failure slices35 min read
Recommend release, rollback, or further testing from a mixed evaluation.
- →Defend a release decision with failure slices
Skills in this course
- 01Locate the first incorrect transition in an agent traceLocate the first incorrect transition in an agent trace.
- 02Design a verifier that measures the requested outcomeDesign a verifier that measures the requested outcome.
- 03Choose an agent architecture from task constraintsChoose an agent architecture from task constraints.
- 04Defend a release decision with failure slicesDefend a release decision with failure slices.
- 05Distinguish tool failure from model decision failureDistinguish tool failure from model decision failure.
- 06Report uncertainty and ownership in an incident answerReport uncertainty and ownership in an incident answer.