The arrivals log
#8am-ai#experiment#signal-harvesting#tooling#timeline
David Olsson
The signal harvester reads every named tool, product, and entity across all 109 meetings and reports when each first and last appeared. Point it at the corpus and it does something the chapters can't: it dates the field's turning points mechanically, from nothing but when a word started showing up.
The names are a clock.
what stayed
Some entities are present the whole way across. They're the bedrock โ the things the group never stopped talking about.
| entity | meetings | first โ last |
|---|---|---|
| Claude | 44 | 2024-08-28 โ 2026-06-24 |
| LLM | 40 | 2024-10-02 โ 2026-06-24 |
| Gemini | 31 | 2024-10-30 โ 2026-06-24 |
| GPT | 23 | 2024-04-17 โ 2026-06-24 |
| OpenAI | 21 | 2024-08-28 โ 2026-06-24 |
| Cursor | 16 | 2024-09-11 โ 2026-06-17 |
Claude appears in 44 of 85 meetings that name anything โ the single most persistent entity in the corpus, present from August 2024 to the most recent Wednesday. GPT is the only thing named in the very first meeting that's still named in the last.
what arrived
The more interesting list is the new words โ entities whose first appearance is 2026. Each one is a turn in the field, dated to the week the group first said it out loud.
- MCP โ first 2025-04-23. The orchestration era. The moment agents could reach tools and data through a standard, the conversation changed from "write a prompt" to "wire a workflow."
- Open Claw / OpenClaw โ first 2026-02-18, then 2026-03-04. The autonomous-swarm inflection. Five meetings in quick succession once it landed.
- Jira โ first 2026-04-08. The agentic layer reaching into real project management.
- Neuralink and Ozempic โ both first 2026-05-27. The philosophical turn, arriving on the same morning. When the conversation moved from tools to the human using them, these two words showed up together.
The harvester didn't know any of that mattered. It counted first-appearances. The narrative fell out of the count.
why the count is the point
Two independent reads of the corpus agree. The hand-written chapters say the orchestration era began in spring 2025 and the swarm turn in early 2026. The mechanical harvester, told nothing, dates MCP to April 2025 and OpenClaw to February 2026. The story and the arithmetic land on the same weeks.
That agreement is the whole argument for the method. If the structure were imposed โ if a writer decided when the turns happened โ the two reads could diverge. They don't. The turning points are in the record, sitting in the data as first-appearance dates, waiting for anything that bothered to read across all of it at once.
This is signal harvesting doing the one thing no single meeting can. Nobody in any room watched MCP go from new word to constant. The harvester sees it because it reads the whole timeline. A tool's arrival is a date, and the dates, lined up, are a history of the field written by counting.