Nate B. Jones answers a question he says he gets constantly: why can Anthropic and OpenAI ship every week while most companies crawl? His answer is that the secret sauce is not AI. It is a set of organizational habits, and he presents them as 15 commandments (written deliberately in mock-biblical "thou shalt" language) for moving repeatable human coordination into code and documents that agents can act on. The core operator job in the AI era, he argues, is to move more of the company's repeatable human interactions toward code, because if every decision has to be re-explained by a person, humans become the rate limit on the business.
Nate opens with the question he keeps getting: why is my business moving slow while Anthropic releases every week? His answer: the secret sauce is not AI. It is that operators at these companies are moving repeatable human interactions toward code. Decisions that used to vanish inside meetings become documents people can refer to and, crucially, that agents can act on. PMs who used to direct engineers through ticketing now work in the terminal alongside them. Reviews become evals. Repeated reminders become systems agents can act on and improve. Design moves into code and into the SDK in ways it never did in the 2010s. He is careful to say not every human interaction moves into software: trust stays human, taste stays human.
Each commandment exists because AI changed the work underneath it. Some move coordination out of meetings and into durable systems. Some move product judgment closer to actual building. Some protect time that quick execution has made more valuable. The rest protect the human relationships that let a company operate at speed with high trust. The single shared intention: find the point of greatest leverage for the business and the humans in it, then move scarce human judgment to that point.
Back when film cost money, a family might take 24 or 48 pictures on vacation. Somebody chose the shot and waited for everyone to look at the camera. The price of film and the complexity of processing supplied discipline. Digital photography removed that barrier, and now you have 40,000 pictures, six versions of the same sunset, and no idea where your kid's birthday went. AI has done this to all of work. The cost of another draft, prototype, analysis, or piece of code is collapsing to zero. Organizations were designed around forced choices; when the forcing disappeared, the choices did not become less important, they just became easier to avoid.
Nate concedes this sounds like every Silicon Valley startup, so he gets specific. Speed is not asking people to work more with AI for the sake of it. Speed is accelerating learning loops, and learning loops are driven by what you can put in front of customers, which is driven by the code you can put together. The old process stretched the distance between bets because every change consumed scarce engineering capacity, and by launch the original question had often moved. Commandment 1 puts every other rule on trial: does this meeting shorten the path from evidence to a better product? Does this document carry a decision? Does this approval actually catch a risk, or is it paperwork? He adds a caveat: distribution is still scarce. Building something nobody sees is still worth zero dollars, and cheap execution raises the premium on deciding what deserves to exist.
The most provocative cluster: product does not make roadmaps (3) and product does not control or direct engineering time (5). On their own those would just produce confusion. But AI has driven the cost of building toward zero, and a team can now put a working version in front of a customer before the old roadmap meeting would have found a free hour on everyone's calendar. That does not remove the need for product judgment; it removes the excuse for product to stay out of the code. So product shall be in the terminal daily (4) and product shall sit with engineering and jam daily (6). The underlying job competencies do not change: engineering still has to answer "does this code work and will it keep working" (a high bar he says teams are using AI to slip past), and product still has to answer "should this exist and does anybody care enough to use it." Same terminal, different questions. Commandment 13 says act like all that matters is one profoundly helpful and delightful customer experience, which is the coordination work the roadmap process used to do. Commandment 12 broadens the design surface: the terminal is a design space, the SDK is a design space, and so is the UI when a payment fails or when an agent hits a permission boundary and has to explain what it needs instead of failing silently. A designer who only designs screens is designing the lobby while the rest of the building goes up without them. Nate expects design to be over-employed, not replaced, because there are many more surfaces needing design input.
No meetings longer than an hour (2). If that forces documentation, that is the point. Documentation as code (11) is meant literally: agents read documents and use them to decide what to do. A document can supply the standard, the source hierarchy, the permissions, the escalation path, the definition of done. Ambiguous documents spread chaos through the system, and that, Nate argues, is the real risk of AI slop. He does not care if you write "you're absolutely right" in a document; he cares whether you know what is in it and understand what you are trying to accomplish. When people sit down to write without that clarity, what comes out reads as slop because the human clarity was never there, and there is no substitute for it. He learned this discipline at Amazon, long before the agent wave, where they typed everything by hand and he once went through 57 drafts of a six-page narrative. They took lunches to review each other's docs, which made everyone's writing better. You do not have to adopt Amazon's exact standards, but in the age of agents you have to care about writing that much. Commandment 7 kills monthly meetings: audit your recurring time commitments and reallocate time to building. Nate mentions he built a meeting-audit tool for this and links it below the video.
This cluster is for the humans. Commandment 8 is about staying flexible and adapting like water over stone. Because teams now communicate more in writing and less face to face, the remaining face-to-face time is more valuable, the writing has to level up, and everyone has to assume best intent. Nate says there is no reason, at any level, to moan or complain: with this much intelligence and tooling available, you can either fix the problem directly or propose a complete fix. The team commandment says you do not build alone. He acknowledges this is controversial given the one-person-billion-dollar-company crowd, but he believes enduring businesses are built with teams, just smaller, well-coordinated ones. One person and one agent can produce an enormous amount of material, but you still need people to reach taste, domain knowledge, customer connection, brand, and the courage to say this does not work. Find a community, his or someone else's. The teaching commandment is about helping people who are moving slower and learning from people who are moving faster, inside organizations and in internet communities alike.
Before closing, Nate names two ways to ruin the whole concept. The first is partial adoption: take the no-roadmap rule but leave PMs out of the code and you have only produced chaos. He sees firms outside the Valley repeatedly failing to realize they have to do this all at once. It is painful, but it is easier to adjust when you set out the whole expectation set. Fifteen rules at once is overwhelming, but the teams he sees successfully accelerating are not biting it off piece by piece; they treat it as an interconnected system where humans, clarity of intent, writing, and how you get into the code all change together. The second failure mode is cherry-picking the comfortable rule ("Nate, you're right, we should get better at no meetings"), which will not get you where you want to go.
The root of why OpenAI and Anthropic ship faster is that they have this culture, teach it, inculcate it, and hire for it, so it reinforces and speeds up over time. Everyone hired at Anthropic already expects that speed, which is far easier than converting a company that expected quarterly releases. That is a massive culture change and it is the thing most people miss. Nate closes by pointing to a Substack article going deeper, the calendar/meeting-audit tool he built, and a companion guide you can feed to your agent to build a complete change-management process for your team. The video, he says, is a starting point for thinking about what it really takes to transform in the age of AI, with a longer discussion coming Monday.