Developer field guide

Keep your assistant sharp in hour three by managing what it can actually see.

The Context Habit — Getting Better Answers from Your AI Assistant

24 pages Screen-friendly A5, prints on A4 and US Letter Read on screen, or print Last reviewed July 2026 Version 2026-07

Somewhere in the third hour of a good session, your AI assistant changes — the model did not get worse; what changed is what it can see, and that is a lever you can learn to manage. This 24-page field guide teaches seven habits for managing your assistant's context: each fits in a single sentence, takes minutes to adopt, and needs no new tool — the same assistant, the same spend, steadier answers across a long session.

Buy now · $11.24

Instant PDF download after checkout on Payhip. Nothing is shipped.

Cover of The Context Habit — Previews show each page at reduced resolution; the download is the full print-quality PDF.

What's inside

What the assistant can actually see — the context window in plain terms: what fills it, who put each piece there, and why your own words are usually the smallest slice
Context rot — the 2025 research that named the third-hour slide, plus four specific symptoms that tell you it is happening in your own session
Reset or continue — a concrete compact-versus-fresh-session rule you can apply immediately, and the ten lines to write down before you do either
The memory file — what earns a permanent line (CLAUDE.md, GEMINI.md, AGENTS.md, rules files), what never belongs there, and the twice-told-correction gate that keeps it honest
Project documents — what lives in the memory file versus the spec versus the conversation, and why pointing the assistant at a file beats pasting it in
Plan first — why agreeing on numbered steps before any code is written is a context habit, not just prudence, and how to read a plan in ninety seconds
Ending a session well — a two-minute exit ritual that removes the ten-to-fifteen minutes of re-explaining you otherwise pay at every cold start
One worked week — every habit applied to one small invoicing project, Monday to Friday, honest about the parts that go wrong
The tools page (July 2026) — the book's generic terms mapped to Claude Code, Cursor, GitHub Copilot, OpenAI Codex CLI, and Gemini CLI: the file each auto-loads, its reset, its compaction, its context meter
A plain-language glossary of every term the book uses

Who it's for

Developers whose long AI sessions slide into re-suggested ideas and forgotten paths
Anyone who wants better output from the assistant they already pay for, without new tools

Who it's not for

Anyone whose sessions run five minutes — the book itself says they never live long enough to degrade
Anyone looking for cost savings — that is the companion book, The Token-Smart Developer; this is the quality side

The context-rot research it draws on is young: the habits are directional, distilled from daily practice and the 2025–2026 literature, not from controlled trials — where your own measurements disagree, trust your measurements. Tool names, commands, and window sizes were last reviewed July 2026 against each vendor's documentation, which wins where a screen has since moved. The worked week is a composite built to teach the method, not a report on a real person or company.

Guidance, not gospel — this is a practical organisation tool, not legal, tax, medical, or religious authority.