Skip to content

Advanced Track Preview

From Delegator to Director

The intermediate track is about becoming a Delegator: someone who defines work clearly enough to hand entire tasks off to AI and be confident the results come back correct. You learned how to decompose work into independently shippable pieces, encode your judgment into skills, verify output with automated tests, and run multiple workstreams in parallel.

The advanced track picks up from there. It is about becoming a Director: someone who designs systems that verify work, respond to failures, and improve over time without you sequencing every step.

What does that shift look like in practice?

  • Evals: Moving beyond "does the code work?" to "is the AI-generated output correct?" Tests verify behavior. Evals verify that the system's output accurately conveys what it should.
  • Observability: Structured logging that lets you (and the system) diagnose failures from data, not guesswork. When something goes wrong, the traces tell you where and why.
  • Autonomous pipelines: CI failures auto-create work items. Agents pick them up in isolated worktrees, implement fixes, and run quality gates, all without you as the middleman.
  • The autonomy slider: Deciding where humans stay in the loop. Not everything should be automated. Not everything needs your direct supervision. The skill is knowing which is which.

The shift is subtle but powerful. A Delegator verifies work: you hand off a task, review the output, confirm it meets your criteria. A Director designs systems that verify work: you define what "correct" looks like, build the pipeline that enforces it, and review the system's output rather than every individual task.

What's in This Preview

These pages introduce two advanced track concepts you can start thinking about right now:

  1. From Tests to Evals: How to check whether AI-generated output is actually correct, not just whether the code runs
  2. The Self-Running System: How CI failures, work items, isolated agents, and quality gates combine into a pipeline that fixes problems without you sequencing every step

These are self-contained previews, not the full advanced lessons. They give you enough to understand the concepts and try them on whatever you are building. If you want the complete experience (observability, golden datasets, the autonomy slider, and hands-on pipeline construction), that is what the full advanced track is for.