Engineer → AI-coding fluency
For: An engineer who can already code but wants to use AI as a serious coding collaborator, not just a fancy autocomplete.
You will be able to: You can run agentic coding workflows, ship faster without shipping more bugs, and you know how to ground AI output in your codebase instead of letting it invent things.
Steps in order
- 01
How large language models work
The mental model that fixes most prompting confusion — prediction, training, inference, why hallucinations happen, why prompt phrasing matters so much. For the operator who wants to understand the mechanism.
— Foundation first. Most engineering mistakes with AI come from missing this model.
10 min·/learn/ai/foundations - 02
Context windows explained: what they limit, what they do not
How much text a model can consider at once, what counts against the window, why quality degrades long before you hit the limit (the "lost in the middle" effect), and how to budget context in production.
— Why your large-context prompts degrade and how to budget around it.
8 min·/learn/ai/foundations - 03
How AI agents work (and where they break)
The minimum that makes something an agent (LLM + tools + loop). What agents are good for, the six predictable failure modes, the autonomy spectrum, multi-agent vs single, and what to log in production.
— The mechanism behind every modern coding agent. The failure modes section is the load-bearing part.
10 min·/learn/ai/foundations - 04
Structuring prompts with XML, roles, and sections
Why <tags> work better than plain text for non-trivial prompts. How role assignments steer behavior. The hierarchy that makes long, mixed-input prompts work — context, instructions, examples, input.
— Long-prompt structure that survives the codebase complexity you will throw at it.
7 min·/learn/ai/prompt-craft - 05
Vibe coding: prompting your way to a working app
The describe to generate to test to refine loop. What "vibe coding" actually means in practice. Where it breaks.
— The starting point — even if you skip the vibe-coding stage, the patterns transfer.
6 min·/learn/ai/build-with-ai - 06
Working with AI on a codebase you didn't write
CLAUDE.md, context files, repo conventions. How to teach an agent your codebase so it stops hallucinating its way through your patterns.
— The realistic case. Almost no one is greenfielding — most work is on existing code.
7 min·/learn/ai/build-with-ai - 07
Catching AI-generated bugs before they ship
Hallucinated APIs, missing edge cases, security holes. How to review AI code without being a senior engineer.
— The patterns AI gets wrong by default. Read before you ship anything AI-generated to prod.
7 min·/learn/ai/build-with-ai - 08
Prompts vs Skills vs Workflows vs Agents — when to use which
Four ways to use AI: a one-shot prompt, an installable skill, a step-by-step workflow, or a multi-step agent. What each one does well, where each breaks, and how to choose.
— The decision framework for what shape your AI helper should take.
8 min·/learn/ai/build-with-ai - 09
How to evaluate an LLM feature is working (without fooling yourself)
Why "looks good" is not evaluation. Building a small eval set (20 cases beats 200), the four grading methods (programmatic, reference, LLM-as-judge, human), what to measure, and how to spot production drift.
— For any AI feature that ships, this is how you avoid silent degradation.
9 min·/learn/ai/foundations
When you finish this path
You can run agentic coding workflows, ship faster without shipping more bugs, and you know how to ground AI output in your codebase instead of letting it invent things. For the next step, browse other paths or the full library.