An opinionated operating system for a four-person AI company.

Why

Writing scales better than talking. A written update can be read by ten people without repeating yourself ten times. It creates memory, exposes fuzzy thinking, and makes meetings dramatically shorter because the context already exists.

How

Write decisions, plans, blockers, customer feedback, specs, and weekly updates by default. Before any important meeting, share the doc first. If the issue can be resolved in comments, cancel the meeting. If a meeting is still needed, use it only to decide.

Trade-off we accept

This can feel intense, less social, and less forgiving for people who prefer to think out loud. We accept that because clarity beats charisma and durable writing beats repeated explanation.

World-class example

Amazon is famous for narrative memos and silent reading at the start of meetings. The lesson is not “write long documents”; it is “clear thinking before talking.”

Sources

Why

A 4-person startup does not win by having better intentions. It wins by compressing the loop from idea to evidence. Daily shipping turns opinions into feedback and stops perfectionism from disguising itself as quality.

How

Ship one meaningful thing every day: prompt improvement, product fix, customer insight, landing page change, sales asset, internal tool, or experiment. Small is fine. Invisible is not.

Trade-off we accept

Some work will look rough, uneven, or unfinished at first. We accept that because polished stagnation is worse than imperfect momentum.

World-class example

Facebook’s early “move fast” culture emphasized building and learning faster instead of protecting the team from every mistake.

Sources

Why

The market does not care how smart our internal reasoning sounds. In a smaller market, being wrong is expensive. The fastest way to stop building fiction is to watch real behavior, hear real objections, and learn why people pay or walk away.

How

Talk to users every week. Watch the workflow, ask what is still manual, ask what they do not trust, ask what almost made them say no, and ask what would make this worth paying for now.

Trade-off we accept

This is messy, repetitive, and occasionally discouraging. We accept that because market truth is more useful than protecting our ego.

World-class example

Sam Altman tells founders to focus on writing code, talking to users, and building the company. Paul Graham’s “do things that don’t scale” is also about staying close to early users.

Sources

Why

In an AI-native company, it is irresponsible to ignore cheap intelligence and consume scarce human attention first. AI should be your first pass for drafting, research, debugging, comparison, synthesis, and sharpening the actual question.

How

Start with AI. Ask it for a first draft, better structure, missing edge cases, likely bugs, alternative approaches, and a clearer explanation. Then bring the team the distilled problem, not the raw confusion. Keep asking: how can I become more productive with AI?

Trade-off we accept

This can produce wrong answers, overconfidence, and occasionally weird artifacts. We accept that risk because the answer is not to avoid AI; it is to verify faster and use humans where they matter most.

World-class example

Sam Altman’s recent writing frames intelligence as becoming extremely cheap; the cultural implication is that every team member should use abundant intelligence before consuming scarce human attention.

Sources

Why

Repetition is a tax on a small team. Every manual copy-paste, recurring checklist, follow-up, formatting task, or rote workflow steals time from product, customers, and leverage. The default question should be: why is a human still doing this?

How

Automate the boring parts with AI, scripts, templates, agents, and simple internal tools. Turn recurring prompts into workflows. Turn recurring workflows into software. If a task keeps showing up, treat automation as part of the job, not a side quest.

Trade-off we accept

We will sometimes spend extra time upfront building tools instead of brute-forcing the immediate task. We accept that because leverage compounds and manual heroics do not.

World-class example

Stripe is known for turning operating patterns into explicit systems, and the best AI-native teams increasingly convert repeatable internal work into tools instead of rituals.

Sources

Why

Unclear ownership is where deadlines go to die. In a team this small, every meaningful outcome needs one person who feels the weight of making it real, not a group chat full of partial concern.

How

For every project, name the owner, expected output, deadline, success metric, and next action. Collaboration is welcome. Diffused accountability is not.

Trade-off we accept

This can feel unequal, exposing, and high-pressure. We accept that because real ownership is more honest than pretending everyone owns everything equally.

World-class example

Amazon’s official leadership principles include “Ownership”: leaders act on behalf of the entire company, not just their own team.

Sources

Why

Comfort creates drag. It shows up as waiting for perfect specs, perfect confidence, perfect consensus, or a prettier version of the same delay. Startups need fast loops and honest exposure to reality.

How

Show rough demos early, launch small tests, ask sharper questions, and kill weak ideas quickly. Speed does not mean sloppy legal work or reckless AI quality. It means shortening the time between assumption and proof.

Trade-off we accept

This can feel uncomfortable, blunt, and occasionally less polished than people prefer. We accept that because comfort is not the product.

World-class example

The Lean Startup frames startup progress as validated learning, not internal activity. Fast cycles beat long hidden work.

Sources

Why

Teams waste time arguing because they never define what would prove the point. For AI products, vibes are especially dangerous. Something can feel magical and still be inaccurate, untrusted, or commercially irrelevant.

How

Track WAU, paid users, MRR, demo-to-payment rate, retention, answer usefulness, citation correctness, hallucination rate, and user trust. When debating, ask which behavior or metric would settle it.

Trade-off we accept

Metrics can oversimplify and create pressure to chase what is easy to count. We accept that because being slightly reductive is better than being beautifully delusional.

World-class example

Lean Startup’s build-measure-learn loop exists because startups operate under uncertainty; progress must be shown by learning from real-world tests.

Sources

Why

Polite silence produces weak decisions. Ongoing resistance after a decision produces slow execution. We want the useful part of conflict, not the lingering part.

How

Disagree clearly, respectfully, and with evidence. Say the hard thing while it can still change the outcome. Once the decision is made, stop relitigating it unless new facts appear.

Trade-off we accept

This can feel more confrontational than consensus-driven cultures. We accept that because fake harmony is expensive and clean commitment is powerful.

World-class example

Amazon’s “Have Backbone; Disagree and Commit” says leaders should challenge decisions respectfully, then commit wholly once the decision is determined.

Sources

Why

A tiny AI startup gets stronger when learning compounds instead of staying trapped in one person’s head. If one person discovers a better prompt, objection handling pattern, or bug fix, the whole team should get faster.

How

Share useful learnings in a short written note: what happened, why it matters, what should change, and what others should try next. Then engage with it by improving, testing, or applying it.

Trade-off we accept

This adds a little documentation overhead and may feel repetitive. We accept that because repeated rediscovery is a far worse use of time.

World-class example

Amazon’s leadership principles include “Learn and Be Curious.” Stripe also treats operating principles as explicit patterns so everyone can benefit from the most effective behaviors.

Sources