Agent Craft
Why?
Learning to build, measure, and improve coding agents.
Goal
By the end of this series you will be able to treat coding agents as junior engineers you can reliably manage: write clear specs, systematically evaluate their output, close the improvement loop, and ship better results with less rework. You move from "prompting and hoping" to deliberate, measurable agent craft.
What You'll Learn
This 12-day series is organized into 4 phases, each building on the last:
Phase 1: Mindset Shift (Days 1–3)
Understand why traditional prompting is insufficient and what replaces it.
- Day 1: Why "just prompting" is not enough — Coding Agent vs ChatGPT + Evaluation-Driven Development
- Day 2: Spec-driven development (why precision replaces ambiguity) — Specs beat prompts (Coming Soon)
- Day 3: Context is the real bottleneck — Context engineering (Coming Soon)
Phase 2: Core Techniques (Days 4–7)
Build the tools to measure and improve agent output systematically.
- Day 4: Measuring what matters — Deterministic checks vs LLM-as-a-Judge (Coming Soon)
- Day 5: Building your first eval harness — Evaluation harnesses and test cases (Coming Soon)
- Day 6: The closed loop — Run → Measure → Improve → Re-measure (Coming Soon)
- Day 7: Failure modes of coding agents — Agent failure analysis (Coming Soon)
Phase 3: Systems & Production (Days 8–10)
Scale evaluation to real projects and production constraints.
- Day 8: Agent autonomy: setting bounds, not paths — Steering without micromanaging (Coming Soon)
- Day 9: From code quality to agent quality — Evaluating the agent's process, not just the final code (Coming Soon)
- Day 10: Production thinking for agents — Observability, cost, reliability, and regression testing (Coming Soon)
Phase 4: Application & Habit (Days 11–12)
Apply what you've learned to your real project and make it stick.
- Day 11: Using your real project as the laboratory — Project-based agent evaluation (Coming Soon)
- Day 12: Building the habit — Daily agent craft practice (Coming Soon)
Why This Series?
Most people start with coding agents and get stuck on prompting. They tweak the prompt, run it again, hope it works better. This is slow and unpredictable.
This series teaches you the other path: Evaluation-Driven Development for Agents. It's the same mindset that turned software engineering from ad-hoc to systematic — you measure what matters, close the loop between observation and improvement, and compound your knowledge over time.
By the end, you'll have the skills that distinguish the "AI Engineer" role from "person who writes prompts." You'll move from being a passenger watching the agent work to being the engineer managing it.
Ready to start? Begin with Day 1: Why "just prompting" is not enough.