ChatGPT Work – Multi-Step Deliverables, Delegation, and Supervision
Lesson 3: Monitoring, Recovery, and Auditability
Lesson Objectives
By the end of this lesson, students should be able to:
- Read progress updates and detect drift
- Design checkpoints and partial-save behavior
- Recover safely from tool failure or interrupted runs
- Create useful execution records without exposing secrets
Lesson Content
Long-running work needs checkpoints. Require the agent to report the current objective, completed steps, sources used, decisions made, pending approvals, and blockers. Progress updates should reveal enough to steer the work without exposing hidden private reasoning.
Failures are often partial: a file was created but not uploaded, a form was filled but not submitted, or one of several records was updated. Recovery begins by inspecting current state rather than rerunning blindly. Use unique identifiers, dry runs, and state checks to avoid duplicate side effects.
Audit records should capture inputs, tools invoked, external sources, material transformations, approvals, outputs, errors, and final verification. They should not record credentials or unnecessary private data. For regulated or high-stakes work, follow organizational logging and retention requirements.
Enroll to continue this lesson.
The preview above shows the lesson objectives and opening lesson content. Enroll to view the full lesson, complete the practice work, and take the lesson quiz.