Series: Overview (you are here) | Process & Methodology | Engineering Thinking | Meta-Skills | Glossary

AI can write code. But software engineers still get dramatically better results from AI coding tools than everyone else. The gap isn’t about typing code faster — it’s a set of mental models that determine whether what the AI writes actually works, scales, and can be maintained.

This series extracts the concepts that transfer directly to AI orchestration and skips the rest. It’s written for people who will never be software engineers but want to direct AI effectively.

The Series

Part 1: Process & Methodology

The 9 highest-leverage concepts. These change how you structure every AI session.

  1. Decomposition — break vague goals into small, testable steps
  2. Iteration over specification — small change → verify → next
  3. Scope management / MVP — build the smallest thing that validates the approach
  4. Checkpoint discipline — save working states, rewind when needed
  5. Acceptance criteria — define “done” before you start building
  6. Fixed time, variable scope — set a time budget, adjust ambition to fit
  7. Spikes — time-boxed investigation before committing to an approach
  8. Vertical slicing — build thin end-to-end, not one layer at a time
  9. WIP limits — finish things before starting new things

Part 2: Engineering Thinking

5 concepts that help you understand what the AI is building and evaluate whether it’s any good.

  1. Technical debt — throwaway vs permanent; why shortcuts compound
  2. Separation of concerns — keep things that change for different reasons apart
  3. State and data flow — where data lives, how it moves, why things break
  4. Debugging as a discipline — systematic narrowing, not guessing
  5. AI reliability calibration — when to trust the AI and when to verify

Part 3: Meta-Skills

3 cross-cutting skills that improve everything else.

  1. Post-incident learning — capture the mechanism, not just the fix
  2. Context management — the AI’s attention is finite; manage it like a scarce resource
  3. Prompt-as-spec — prompts that function like engineering specs, not feature requests

Glossary

~40 terms you’ll encounter during AI coding sessions, explained in plain English. From “what is a terminal?” to “what is yak-shaving?” — grouped by theme, starting with the absolute basics.

What This Isn’t

This is not a “learn to code” guide — AI can write the code. This is not a tool tutorial — other courses cover that well. This is the gap between the two: how to think about software well enough to direct AI that builds it for you.