Vibe coding, properly.
25 guides on building with AI coding agents — what works, what breaks, and what to do when it does. Written for people shipping, not for search engines.
Fundamentals
Context windows explained for people who just want to ship
Why your agent gets worse an hour in, the signals that tell you it has happened, and the handful of habits that keep sessions sharp.
6 min read
What vibe coding actually is, and where it breaks down
A working definition of the loop, the six places it reliably fails, and a one-sentence test for whether a task is safe to hand over.
6 min read
Why your agent forgets everything, and what to do about it
Three tiers of memory for coding agents, and why a lint rule enforces your standards better than any paragraph of markdown ever will.
6 min read
Workflow
Shipping to production what an agent built
Agent code is correct on the happy path and thin everywhere else, so shipping means checking the parts nobody prompted for.
6 min read
Structuring a repo so an agent can navigate it
Six concrete changes that stop your agent burning nine tool calls working out where the routes live.
6 min read
The retry loop: spotting it and breaking it
Two attempts on the same error is the whole budget, and here is exactly what to do when the second one fails.
7 min read
When to stop the agent and write it yourself
Six signals that another prompt is wasted, and the skeleton-and-fill pattern that beats both retrying and taking over completely.
6 min read
Writing a prompt that produces working code the first time
The five parts of a prompt that lands, why pointing at an existing file beats describing a pattern, and a template you can paste today.
6 min read
Quality
Debugging an AI-generated codebase you have never read
A repeatable order of operations for fixing code you never wrote, from reproducing on demand to finding the seams where generated code goes quiet.
6 min read
How to review code you did not write
An ordered pass for large agent-written diffs, so your attention lands on the parts that can actually hurt you.
6 min read
Security mistakes agents make repeatedly
The same short list turns up in agent-written code every time, and all of it passes your tests. Here is the list, with the fixes.
7 min read
Tests are how you go faster with an agent, not slower
Tests are what let the agent self-correct without you in the loop, turning ten human turns into one review.
6 min read
The duplicate-code problem nobody warns you about
Three copies of the same helper is not untidiness, it is a bug you fix once and ship twice. How to find them and stop making more.
6 min read
Optimisation
Choosing a Model for the Task: When Cheap and Fast Wins
A task-by-task guide to when a cheap fast model wins, when it costs you an hour of cleanup, and how to test the trade-off on your own repo.
6 min read
Cutting Your Token Bill Without Cutting Quality
Most token waste is context hygiene, not model choice. Where the spend actually goes, and the session habits that cut it without hurting output.
6 min read
Making an Agent Work on a Large Monorepo
Agents that shine on a 40-file project invent a map when the repo has 4,000 packages. The fix is navigation aids and scope, not a bigger window.
6 min read
Measuring Whether the Agent Is Actually Making You Faster
Agents feel fast, and feeling fast is not the same as being fast. A two-week log, honest arithmetic, and why rework rate is the number that matters.
6 min read
Tools
Building a Reusable Agent Setup You Actually Carry Between Projects
Stop re-teaching the agent your commit format on every new repo. Build a personal layer that travels with you, kept clean of the client's project layer.
6 min read
Rules, Skills and Hooks: Which One for Which Job
Rules are a request, hooks are a guarantee, skills are the procedure in between. A decision table and three worked examples for choosing.
6 min read
The Tools Worth Paying For, and the Ones That Are Not
A category-by-category verdict on the subscriptions in your stack, plus the cancel test and a worked monthly budget for a solo developer.
6 min read
Business
Building a Portfolio When the Agent Wrote Half the Code
Your GitHub profile stopped proving you can code. What a portfolio has to demonstrate now, and how to restructure yours around it.
6 min read
Going Freelance as a Vibe Coder: What Clients Actually Buy
You can build a working product in a weekend now. That is not the thing anyone is paying you for, and confusing the two will cost you.
6 min read
Handing Over an AI-Built Codebase
Handover is where you find out how well you know code you did not type. The audit, the documentation, and the disclosure you owe the client.
6 min read
Pricing AI-Assisted Work Without Underselling Yourself
If you bill by the hour, every efficiency gain from your agent goes straight into the client's pocket. Here is how to stop giving it away.
6 min read
Writing a Proposal for Work You Will Build With an Agent
You can credibly promise in two weeks what the agency quoted at six. Write that proposal the old way and you get disbelieved, or trapped.
6 min read