Writing
Writing
Essays
- 2026-09-30 What a language model doesn't see The first drafts I get from language models present in text when people need to see the data, and on hard technical work they stop writing for the reader. I think it's one problem, and I'd like to work on it.
- 2026-09-29 Questions I'm still working on I build almost everything with AI now, and I train machine learning models for weather. Both have changed how I work, and both have left me with questions I haven't settled. Some are practical and some are ethical. I ask them about my own work first, and then about what they mean for other people.
- 2026-09-27 Coding agents without locks How three vendors' coding agents edit the same code concurrently without stepping on each other.
- 2026-09-27 A grounding doctrine for an agent's memory Provenance or it isn't evidence; re-ground before acting; modality survives every hop.
- 2026-09-27 Lint for LLM prose Thirteen rules that took 87 chapters from 61 failing to zero.
Field notes
- 2026-08-27 The tool catalog that was never sent I was told tool definitions were costing 65,811 characters on every call.
- 2026-08-26 160 places my agent could lose your words One truncated safety instruction started a program.
- 2026-09-11 93% of the bill was history I assumed tool calls were flooding my agent's context.
- 2026-07-26 Putting the right things in front of the agent, every turn I moved what the model sees each turn into plain code, and tested it blind.
- 2026-09-10 When the prompt files fight each other An agent's standing files compete with each other and with fresh memory for the same space.
- 2026-09-02 A 424-byte question that cost 4.2 million tokens The budget that should have stopped a runaway delegation had been recording zero for eight days.
- 2026-07-21 Five fixes that killed nothing A subagent ran for forty minutes and 117 browser calls, invisible to the agent that started it.
- 2026-09-24 The paper that sounded like a model talking to itself My paper drafts were accurate and read wrong.