Articles
Longer pieces on software craft and on working with AI coding agents, newest first. Each one links to the tenets it draws on.
Agentic coding with checks an agent can't skip
Agentic coding is an AI agent working through a task in a loop. Here's a loop that keeps its code maintainable, with a check at each step and what it costs.
AGENTS.md contents and where agents stop reading
AGENTS.md holds the Markdown instructions most coding agents load. Here's what to put in it, where agents stop reading and an example to copy.
AI code review checklist with one line per rule
Review code with AI against a checklist of 51 named standards, one line each, with review notes to paste, a worked example and a chat prompt.
AI token cost and why messy code costs more
Coding agents pay for every token they read, and messy code takes more reading. What the studies found, where splitting code cost more and what to do about it.
Clean code principles a linter can enforce
Clean code principles from Robert C. Martin's book, split into the ones a linter enforces in five languages and the ones only a reviewer can judge.
Code complexity limits and how linters count them
Code complexity is measured as cyclomatic or cognitive complexity. How each is counted, why linters score one function differently and what limits to set.
Code smells a linter catches and how to fix them
Seven common code smells, from long functions to magic numbers, each with its usual fix and its lint rule in five languages, plus the smells left to review.
Coding standards that the build enforces
Coding standards are the rules your code has to meet. Write them as lint limits and CI checks that fail the build, so people and coding agents keep to them.
Context engineering and what a coding agent loads
Context engineering splits a coding agent's rules between every session, on demand and the linter, with token costs measured in Claude Code.
Copilot instructions and what each surface reads
GitHub Copilot reads .github/copilot-instructions.md almost everywhere. Here's what each Copilot surface reads, what code review reads and examples to copy.