Jeff McMillan has been on Invisible Machines three times—each visit a different chapter of the same argument: you cannot scale intelligence on vibes; you scale it on knowledge, evals, and humans who have something to lose.
2022 — Early Mover (S2E22)
Pre-ChatGPT, McMillan described Morgan Stanley’s OpenAI partnership as a compliance-aware experiment in curated content—not a race to ship a demo. Leadership buy-in, legal review from day one, hackathons, and prompt literacy for advisors established a pattern: the model attracts budget; the governed corpus earns trust. Read the S2E22 summary.
2024 — Operational Curation (S3E13)
With David Wu, McMillan made the work concrete: sixty thousand documents toward twenty thousand with domain owners, an internal Wikipedia mental model, vector-plus-metadata hybrid retrieval, regression testing when models change, side-by-side human vs AI comparison, and prompt coaching in production. This is the boring middle enterprises skip when they buy orchestration instead. Read the S3E13 summary.
2026 — Agent-Scale Checklist (S7E12)
Now founder of McMillanAI, he names what board decks omit: data accessibility near one hundred percent, semantic structure, embedded controls, custom evals that survive upgrades, process maps honest enough for consultants, and agent-in-the-loop accountability when machines act at reach. Five agents can be brute-forced; fifteen thousand cannot. Humans are not the bottleneck—they are the accountability layer that makes scale survivable. Read the S7E12 companion essay or the full transcript.
One Through-Line
Three conversations, one stack: accessible information, semantic connective tissue, control and monitoring—then models, orchestration, applications. McMillan lived the multi-year version inside a global institution; Robb Wilson’s lawn-mower line names the mistake of fighting over apps while the foundation starves.
If your organization is pursuing agentic AI without that foundation, this cluster is the checklist your deck is probably missing.