Two years of 8am AI

A group has met most Wednesday mornings at 8am since April 2024. The subject is whatever is actually changing in the work as the models improve. Every session was recorded.

Two years in, that's 109 meetings. The transcripts sat in a folder doing nothing until a pipeline read across all of them at once. What came back: fourteen recurring themes and about 760 distinct ideas. None of them planned. All of them mined from what was said.

This is the first read of the whole corpus, end to end.

It started as "8am AI School" — numbered workshops with agendas. By mid-2025 it was just "8am AI!", a standing conversation. The name change tracks a real shift in what the meeting was for.

---

the clearest arc is agents

Follow one thread across two years and follow the agentic one. It's the largest cluster in the corpus — 143 ideas — and the only one with a clean trajectory.

In 2024, agents are built by hand. David calls them "scrubs": a basic set of instructions that makes an LLM do one job. One agent, one prompt, one task.

In 2025, the work is orchestration. N8N, MCP servers, workflow builders. The question moves from "make an agent" to "wire agents together."

In 2026, the agentic layer goes everywhere, then OpenClaw and autonomous swarms arrive. People describe directing fleets of agents instead of doing the work.

From writing one prompt to directing a swarm. The group lived the whole curve in public.

---

the threads that carried the most weight

Agentic workflow leads with 143 ideas. Behind it:

Context engineering, 97. The model is the commodity. The context is the lever. It starts as "scrubs" in 2024 and ends as corpus-building and one question — "how do you know" — in 2026.

Product and business, 61. The turn from selling AI as a product to selling it as a process. Market access, not capability, is the scarce thing.

Developer productivity, 59. A log of tool churn — GPT-4, Claude-in-Canada, N8N, Cursor, slop code, multi-agent — kept with its failures. License violations. Context corruption. "You're absolutely right" as a tell.

System-vs-design thinking, 59, and meeting-cadence, 54, round out the top.

---

what the corpus is for

Three things fall out of two years of record that no single meeting gives you.

Trajectory. You can see which ideas were early. David ran collaborating agents against real projects in mid-2024, a year before "agentic" was a headline. The corpus is the receipt.

Recurrence. Themes that return without anyone deciding to raise them — context engineering, atomic-to-architecture — are the ones worth building on. The pipeline surfaces them mechanically.

Provenance. Every idea cites the meeting it came from. The claims here trace back to specific Wednesdays in transcripts/.

---

The book isn't written. It accretes from what was already said. This is the index. The chapters follow from the record.