Enrolment
This course is currently free to enrol in.
Public course
Enterprise AI — Course 12: Evaluating Agent Behaviour
Evaluate agent actions, effects and constraints, challenge misleading evidence and defend a restricted pilot decision.
AI agentsevaluationtool usetracesconstraintspilot readiness
By Telari Labs
Enterprise AICourse 12 of 25
Learning on Telari
Created by Telari Labs, delivered through Telari
The course creator controls the material and assessment standards. Telari provides structured lessons, practice, feedback against those standards and a record of your progress. See how Telari works.
AI access
The first managed lesson is included on Free. Continue with Plus or your own supported API key; premium assessments and certificates may require Plus.
Audio
No reviewed lesson audio is currently advertised for this course.
Lesson list
- Outcome and trajectory evidenceYou should judge a task result and the route used to obtain it separately, then apply the supplied acceptance rule.
- Tool use and actual effectsYou should locate the precise demonstrated failure in a tool trace without treating every failed task as the same capability failure.
- Controlled environments and resetYou should decide whether a trial provides valid evidence of the agent capability being tested.
- Exceptions and interruptionsYou should design and interpret discriminating cases for an agent facing denial, failure, interruption or conflicting content.
- Judges and independent recordsYou should challenge an evaluation judgement using the right execution evidence and a calibrated assessment of grader limits.
- Constraints, cost and human effortYou should compare candidate agents using separate outcome, constraint and resource evidence.
- Restricted release judgementYou should recommend whether a bounded agent is ready for a specified restricted pilot and explain the remaining uncertainty.
- Restricted Pilot Evidence ReviewYou should apply the course capabilities to fresh evidence and defend a bounded pilot recommendation.