Nate's operating system for moving a company from tool adoption to real AI-native speed. They are written as literal “thou shalt / thou shalt not” commandments, grouped into clusters — and his central warning is that they only work as a single interconnected system. Adopt one in isolation and you produce chaos, not speed.
| # | Commandment | Cluster |
|---|---|---|
| 1 | Thou shalt not slow down engineering — thou shalt go fast. Speed = accelerating learning loops, not busywork with AI. | Speed |
| 2 | No meetings longer than an hour. If that forces documentation, that's the point. | Writing |
| 3 | Product does not make roadmaps. A working version can reach a customer before the roadmap meeting finds a free hour. | Product |
| 4 | Product shall be in the terminal daily. You get into the code one way or another. | Product |
| 5 | Product does not control or direct engineering time. | Product |
| 6 | Product shall sit with engineering and jam daily. Touch the material, bring the customer, decide while it's becoming real. | Product |
| 7 | Kill monthly meetings. Audit recurring time commitments; reallocate to building. | Writing |
| 8 | Stay flexible — deliver value like water over stone. Adapt as the work shifts. | Human |
| 9 | Assume best intent — thou shalt not moan or complain. (behavioral cluster; not numbered on-screen — see note) | Human |
| 10 | Either directly fix the problem or propose a complete fix. Everyone has the intelligence and tooling to do so. (inferred — see note) | Human |
| 11 | Documentation as code. Agents read docs to decide what to do; ambiguous docs spread chaos downstream. | Writing |
| 12 | Design moves into the code. The terminal, the SDK, and the UI are all design spaces — including agent experiences. | Product |
| 13 | Act like all that matters is one profoundly helpful and delightful customer experience. | Product |
| 14 | Build with a team — you do not build alone. Small, well-coordinated teams beat the solo “one-person unicorn.” | Human |
| 15 | The teaching commandment. Help those going slower; learn from those going faster. That's how everyone gets faster together. | Human |
The gap between Anthropic/OpenAI and everyone else isn't better AI — it's that AI-native teams have moved their repeatable coordination out of meetings and documents and into working code that agents can act on. Nate codifies this as 15 “commandments” spanning speed, product, writing, and human culture, and insists they only work adopted all at once as one system — partial adoption (e.g., killing the roadmap without putting PMs in the terminal) just manufactures chaos.
Nate opens by rejecting the common explanation for why Anthropic and OpenAI ship weekly while most companies crawl: it is not that their AI is better. The distinguishing factor is what they rebuilt underneath the model — moving repeatable coordination out of meetings and documents and into code that agents can operate against. Trust and taste stay human; the repeatable coordination does not.
He wrote 15 commandments (with deliberate “thou shalt not” flavor) to guide organizational change. Each exists because AI changed the work beneath it: some move coordination out of meetings into durable systems, some move product judgment closer to building, some protect newly-valuable time, and the rest protect the human relationships that let a company run at speed with high trust. The single unifying intent: find the point of greatest leverage and move scarce human judgment to that point.
When film cost money, its price imposed discipline — you took 24 shots, someone chose the moment. Digital removed that barrier and gave you 40,000 photos and six versions of the same sunset. AI has done this to all knowledge work. The collapse in the cost of another draft didn't make choices less important — it made them easier to dodge. Hence the two questions every AI-heavy company must answer.
“Go fast” isn't Valley platitude. Speed means shortening the loop from evidence to a better product. Commandment 1 puts every other rule on trial: does this meeting/document/approval/handoff shorten the learning loop or catch a real risk? If you can't name the job it does in an agentified pipeline, it's slowing you down. The other scarce thing remains distribution — cheap execution raises the premium on deciding what deserves to exist and getting it in front of the right person.
The most provocative rules: product doesn't make roadmaps (3) and doesn't direct engineering time (5). Stated alone they'd just create confusion — but because AI drove build cost to zero, a team can ship a working version before the old roadmap meeting even convenes. This doesn't remove product judgment; it removes product's license to stay out of the code. So PMs must be in the terminal daily (4) and jam with engineering daily (6), owning the coordination the roadmap process used to do — all in service of one delightful customer experience (13). Design, too, moves into the code (12): the terminal, SDK, and UI are design surfaces, including agent experiences (permission boundaries, failure fallbacks). A designer who only designs screens is “designing the lobby while the rest of the building gets built without them.” Nate expects design to be over-employed, not obsolete.
No meeting over an hour (2) — and if that forces documentation, good, because agents need written feedback. Documentation-as-code (11) is meant literally: agents read docs to decide what to do, so a doc supplies the standard, source hierarchy, permissions, escalation path, and definition of done. Ambiguous docs push chaos downstream — the true source of “AI slop” is missing human clarity of intent. Nate learned this discipline at Amazon (57 drafts of a six-page narrative; lunchtime doc reviews) and argues that document culture must survive into the agent age. Commandment 7 kills monthly meetings and audits recurring time so more of it goes to building — he built a meeting-audit tool for exactly this.
This cluster makes AI's speed survivable for human teams. Stay flexible like water over stone (8). Assume best intent and don't complain (9) — everyone has enough intelligence and tooling to fix or fully propose a fix (10). Build with a team (14): Nate rejects the one-person-billion-dollar-company thesis; enduring businesses come from small, well-coordinated teams that catch each other's mistakes, clarify intent for AI, and supply taste, domain knowledge, and the courage to say “this doesn't work.” Find a community. And the teaching commandment (15): help those moving slower, learn from those moving faster — that's how everyone accelerates together.
Partial adoption — take “no roadmaps” without putting PMs in the code and you just manufacture chaos. Teams outside the Valley often don't realize they must adopt the whole system at once. Not treating it as culture change — Anthropic and OpenAI hire, teach, and inculcate this culture, so it compounds; every new hire already expects this speed, which is far easier than converting a quarterly-release org. The whole business must move together.
"What is their secret sauce? What is the thing that they have? Hint, it's not AI."
"If every decision has to be re-explained by a person, the humans become the rate limit on the business."
"The cost of another draft, another prototype, another analysis, another piece of code is collapsing to zero... the choices did not become less important. They actually just became easier to avoid."
"A designer who only designs screens is now designing the lobby while the rest of the building gets built without them."
"I don't care if you say you're absolutely right in a document. I do care if you know what's in it."
"Cheap individual output is not the same as making something that people want."
| Time | Topic |
|---|---|
| 00:00 | Why AI-native companies ship faster than everyone else |
| 00:47 | The real shift: moving repeatable coordination into code |
| 02:03 | Why I wrote 15 commandments to guide the change |
| 03:09 | The photography analogy: what got cheap, and why judgment matters more |
| 04:22 | Commandment 1: don't slow down engineering, go fast |
| 06:23 | Commandments 3–6: kill the roadmap, get product into the terminal |
| 08:27 | Commandment 12: design moves into the code, the SDK, and the UI |
| 09:39 | Commandments 2 and 11: no meetings over an hour, documentation as code |
| 11:39 | Commandment 7: kill monthly meetings, and the audit tool I built |
| 12:39 | The human cluster: flexibility, best intent, and building with a team |
| 14:37 | Commandment 15: the teaching commandment, and the two ways this fails |
| 17:11 | The companion guide, the meeting tool, and what's next |