Notes | Async DigitalWorking notes on building software with AI agents doing a large share of the work. The subject is the system around the model rather than the model itself, and the code underneath both: how sessions coordinate, where the rules live, and which checks can actually fail.https://async-digital.com/notes/feed.xml2026-09-15T00:00:00ZHolding it righthttps://async-digital.com/notes/holding-it-right/2026-09-15T00:00:00Z2026-09-15T00:00:00ZRob NashOne talk made a weak model do a real job with a harness. The other says no harness fixes what agents do to a codebase. I stopped reading every line, so I checked where my own checks stop.One line, fourteen packageshttps://async-digital.com/notes/one-line-fourteen-packages/2026-09-13T00:00:00Z2026-09-13T00:00:00ZRob NashOne dependency line pulled 14 packages and about 40,000 lines of code into the app. Apple closed most of the gaps and an agent writes most of the boilerplate now, so what does a third-party architecture still buy?One caller hid the crashhttps://async-digital.com/notes/one-caller-hid-the-crash/2026-09-11T00:00:00Z2026-09-11T00:00:00ZRob NashA test suite passed seven out of seven while the code under it corrupted the task allocator on every run. Only one file called the initialiser that triggers it, and the bug needs two.Asking if it's finished strips the lockhttps://async-digital.com/notes/asking-strips-the-lock/2026-08-30T00:00:00Z2026-08-30T00:00:00ZRob NashThe polite way to coordinate two AI sessions sharing one save file, asking whether it has finished, is the message that strips its protection. I caused the bug live while researching it.A correction looks like a duplicatehttps://async-digital.com/notes/a-correction-looks-like-a-duplicate/2026-08-30T00:00:00Z2026-08-30T00:00:00ZRob NashA file my agents keep said the opposite of the truth about my own machine for weeks. A June paper measures why corrections get thrown away by the check meant to stop duplicates.The template wrote the tickethttps://async-digital.com/notes/the-template-wrote-the-ticket/2026-08-20T00:00:00Z2026-08-20T00:00:00ZRob NashAn agent filed nine paragraphs to change two words in a comment. It had followed its instructions exactly.Coordination lives in the harness, not the prompthttps://async-digital.com/notes/coordination-lives-in-the-harness/2026-08-05T00:00:00Z2026-08-05T00:00:00ZRob NashA session hands itself off to the next one, and won’t trample work another session is already doing. Neither behaviour is a prompt.Your rules should travel with youhttps://async-digital.com/notes/your-rules-should-travel-with-you/2026-05-18T00:00:00Z2026-05-18T00:00:00ZRob NashThe rules you build with an AI assistant are professional capital. They belong to the studio, not the vendor.Notes on coordinating AI agent swarmshttps://async-digital.com/notes/coordinating-ai-agent-swarms/2026-05-02T00:00:00Z2026-05-02T00:00:00ZRob NashThe primitives missing from agent-swarm runtimes, and the disciplines that fill the gap.Thirty days of agentic developmenthttps://async-digital.com/notes/thirty-days-of-agentic-development/2026-04-24T00:00:00Z2026-04-24T00:00:00ZRob NashI counted what thirty days with an AI coworker actually produced, because I could not find an honest figure to compare against.