Short articles from Team Cartographer on how people actually build with agents, what session history can teach, and the craft of prompting and correcting. Every claim cites its source.
Run a retrospective on your agent session, not only the codeBy Team Cartographer · November 1st 2026 · 4 minute read Teams have long paused to ask how they worked, not just what they shipped. Here's what Scrum and the Agile principles say, and a small version for one.
Is a coding agent a pair programming partner? What research saysBy Team Cartographer · October 31st 2026 · 4 minute read Pair programming research found it isn't uniformly beneficial. What that meta-analysis and Anthropic's docs suggest about working with an agent.
Your agent's session history expires: what to keep, and whenBy Team Cartographer · October 29th 2026 · 4 minute read Claude Code deletes old transcripts after 30 days by default. What the docs say, and a simple way to keep the record of your build.
Ironies of automation: staying sharp when an agent writes the codeBy Team Cartographer · October 27th 2026 · 4 minute read Bainbridge warned in 1983 that automation can erode the skills needed to supervise it. A 2026 trial of AI-assisted coding suggests a similar worry.
Feeling fast versus being fast: timing your agent-assisted workBy Team Cartographer · October 25th 2026 · 4 minute read Developers in one trial felt faster with AI and measured slower. What that and the planning fallacy suggest about timing your own agent sessions.
Writing a spec before the prompt: what the research saysBy Team Cartographer · October 23rd 2026 · 4 minute read Vague requirements are an old problem, and agents don't remove it. What a requirements survey and Anthropic's docs say about writing the spec first.
When to start a fresh session: long chats with a coding agentBy Team Cartographer · October 21st 2026 · 4 minute read Long agent sessions get heavy. What Anthropic's own docs say about context, /clear, /compact and rewind, and a plain rule for when to start over.
Write down the why: decision records for agent-built codeBy Team Cartographer · October 19th 2026 · 4 minute read Code shows what you chose, not why. What Nygard and Microsoft's guidance say about decision records, and a ten second habit for agent sessions.
Praise is not evidence: appraising an agent's work honestlyBy Team Cartographer · October 17th 2026 · 4 minute read A confident "all done" feels like proof, but it isn't. What research and vendor docs say about checking an agent's work against evidence.
Reading your own dead ends: what abandoned work can teachBy Team Cartographer · October 15th 2026 · 4 minute read Abandoned branches and rewound sessions are the part of a build nobody keeps. Here's why they're worth reading, and what the docs say about keeping them.
Declaring done before you start: goals as a yardstickBy Team Cartographer · October 13th 2026 · 4 minute read Decide what finished looks like before your first prompt. Here's what Scrum and vendor docs say, and what a checkable yardstick holds.
Why a correction is the most informative prompt in a sessionBy Team Cartographer · October 11th 2026 · 4 minute read A correction shows what you meant, what the agent assumed, and where they split. Here's what the research and vendor docs say about it.
What the research says about reviewing code you didn't writeBy Team Cartographer · October 9th 2026 · 4 minute read Two studies of code review put understanding the change first. Here's what they found, and what it means when you didn't write the code.