Free download. No email required. The A3 poster is built to hang above a desk; the A4 print sheet comes straight from this page.
Why this Codex exists
Most AI startups don’t die at the bottom. They die at the peak, every indicator green. The crack opens behind the good numbers and stays fixable for a while — but only if the founder is looking for it. The old SaaS playbook misses it, because in the old world the marginal cost of one more customer was effectively zero. In AI, every customer carries a recurring inference cost, and the same growth that built classic SaaS can now consume an AI company from the inside.
The Codex is the answer to that asymmetry. Not a list of tips. A code by which decisions get checked. It sits on top of the AI Survival Canvas — the 9-block X-ray of an AI business — and translates diagnosis into operating discipline. Fourteen lines. Four layers. One sentence at the bottom that compresses the whole book.
How to use the Codex
- Use it as a filter, not an audit. Before a big decision — a hire, a price change, a fundraising round, an enterprise contract — check whether you are quietly violating any of the fourteen. If you are, owe yourself one causal sentence explaining why this time is different.
- Know your group. Most founders break the same two or three principles across years. Layer-1 founders confuse output with workflow embeddedness. Layer-2 founders confuse revenue with margin. Layer-4 founders walk into investor meetings without a verdict of their own. Find yours, and keep it visible.
- When principles conflict, Principle 9 arbitrates. Treat the layer that is actually leaking. A layer-1 problem cannot be cured with layer-4 money.
Layer 1 — Strategy: will you even reach scale?
Strategy catches the causes of death that surface before economics matters. A perfect unit economics model is worthless if the company never reached the scale where it would have run.
1. Build a wedge, not a sticker
A viral AI product grows easily, because the first try is nearly free — and leaves just as easily. The only thing that turns growth into survival is how painful you are to leave. A deep wedge sinks into the workflow through accumulated context, the right to act, and a team protocol. A surface sticker peels off in a single motion. Jasper hit a $1.5B valuation in 2022 and a 30% revenue-target cut nine months later, because the moment ChatGPT did almost the same for free, leaving cost nothing.
2. Distribution lives in the product, not the ad
What spreads isn’t the landing page. It’s the work output. A product with no built-in loop lives exactly as long as its ad budget. A paid channel stacked on top of variable inference cost is scaling at a loss — every additional click carries a recurring cost the founder pays out of pocket. If growth doesn’t work without marketing dollars, you don’t have a Distribution Engine. You have a pump that stops the day you stop pouring.
3. Retain on M12, not on the first week
The first months of an AI product are tourist noise: tried it, left. Sign-up peaks and week-one DAU prove curiosity, not product–market fit. The true PMF signal is M12 retention divided by M3, cohort over cohort. Stable or rising, the product is alive. Falling, and the growth chart everyone is celebrating is just masking churn one quarter ahead of where the dashboard reports it.
4. The moat is in the workflow, not in the model
In AI, model quality is a slippery surface. Stanford HAI found the inference cost for GPT-3.5-level performance dropped more than 200× in two years, and open-weight models closed the gap with closed ones. Switching providers is one line of code. The only durable moat is embeddedness in how the work gets done — context, actions, system of record, team protocol. To build a real wedge, weaken the false lock-in on a specific model and deepen the real one underneath.
Layer 2 — Economics: will you survive your own success?
Economics is where AI inverts the old SaaS religion. Growth is no longer free; it is the rate at which the gap between revenue and cost is widening. Four principles keep the gap on your side of zero — and the financial model is where you watch it move.
5. Own at least one inference path
A business where every request passes through someone else’s API sells at the vendor’s price. When the vendor raises prices, your margin disappears the same day. Owning an inference path does not mean owning a model. Caching, distillation, local fallback, multi-provider architecture — all count. At least part of the per-request cost has to sit under your control, so a single repricing email does not redefine your business.
6. Price must reflect consumption
A fixed price on top of variable inference cost means the company is subsidizing its heaviest users. In practice, the top 10% of users routinely burn up to 60% of the compute and pay the same as everyone else. Every new “power user” the team celebrates makes the economics worse. Price has to move with usage — seats with limits, credits, tiers, or a pure usage line — otherwise you are paying for someone else’s enthusiasm.
7. Measure channels by gross profit, not revenue
Channels that returned revenue above CAC was the SaaS test. In AI it is only half the equation: every acquired user carries a variable inference cost that scales with use. Judge a channel by what is left after COGS, not by what came in. Otherwise you will think you are growing while the servers quietly eat the rest — and the channels that look most efficient on revenue can be the deepest holes in gross margin.
8. A round is not a business model
Raised capital masks unit economics that don’t work. Every next round buys time to the same verdict, at higher burn and a higher post-money. Runway here is not a safety margin. It is a diagnosis postponed. If the business does not work without a round, it will not work after one either — the round just changes how loudly it doesn’t work, and to whom.
Layer 3 — Discipline: protection from self-deception
Discipline is not a procedure layered on top of the work. It is protection from self-deception. The founder’s main enemy is the founder: the narrative built for investors, the team, and oneself quietly replaces diagnosis. Only one thing guards against that — repeatable, ritual honesty.
9. Fix the layer where the leak is
Startups die because they treat the wrong layer. Fixing CAC when the wedge is broken. Raising a round when retention is dead. Optimizing margin when there is no distribution loop. Bringing in an investor to fix economics that structurally do not work. A layer-1 problem cannot be cured with layer-4 money. A fresh round only delays the verdict — at higher burn. When two principles in this codex conflict, this one arbitrates.
10. Stress-test before every big decision
A big decision: a senior hire, a step-change in marketing spend, a new round, an enterprise contract, a price move. Before each one, run the scenarios — what breaks in the Conservative case, what happens at double inference cost, what is lost at the renewal cliff. The point is not to paralyze yourself. It is to know which part of the business you are about to put at risk. A decision made without a stress test is a blind bet dressed up as strategy.
11. 90 minutes a quarter, 9 minutes a month
A review rhythm is not an extra procedure. It is the founder’s attention operating system. In classic SaaS, a quarterly review was enough. In AI, prices jump, models update, and product–market fit is renegotiated every few months — so quarterly already lags. Full review once a quarter against the AI Survival Canvas. Compressed review once a month against the Codex. An immediate run after any shock — price, product, or competitive.
Layer 4 — Verdict: the language you speak with investors and with yourself
The verdict layer translates strategy, economics, and discipline into one causal sentence — the sentence an investor funds, and the one a founder uses to keep an honest internal scorecard.
12. A causal sentence, not a dashboard
The Survival Verdict is not a metric. It is a three-part causal sentence — now, what breaks at scale, why. “LTV:CAC 2.1, AI Layer GM 34%, runway 11 months” is a report. “This company grows users faster than contribution margin, and at 10× scale inference cost becomes the business itself” is a verdict. The first is forgotten in an hour. The second is quoted at the board.
13. An honest verdict is the first thing the investor hears
An experienced VC delivers a verdict on the company inside the first twenty minutes of the conversation. The only question is whether they hear yours first or invent one along the way. A founder who delivers a causal verdict up front controls the meeting. A founder with twelve metrics and an optimistic narrative hands the investor the right to diagnose the business for them — and the diagnosis usually lands harsher than the one the founder would have offered.
14. Optimize for survival, not growth
Classic strategy frameworks optimize for growth — how to grow faster. When inference cost moves by multiples in months, models update in weeks, and PMF is renegotiated each quarter, a growth framework no longer covers the base task: staying alive. The survival system does. First, don’t break under growth — then grow faster. Not the other way around. In an environment where lasting is harder than launching, survivability stops being defense and becomes the competitive advantage.
When principles conflict
A fourteen-line manifesto can sound as if everything applies at once. It doesn’t. Sometimes survival conflicts with growth. The most common collisions: Principle 6 (price reflects consumption) versus the simple, marketable pricing investors want to see; Principle 8 (a round is not a business model) versus needing a round to reach working economics in the first place; Principle 14 (optimize for survival) versus investor pressure to show growth at any cost.
The rule is simple. In a conflict, the principle closer to the leaking layer wins. Strategy is leaking — pick the strategic principle. Economics is leaking — pick the economic one. Treating a lower-layer leak with a higher-layer cure is the most common founder mistake, and Principle 9 — fix the layer where the leak is — is the chief arbiter.
The Codex and the AI Survival Canvas
The Codex doesn’t stand alone. It is the operating layer on top of the AI Survival Canvas — a one-page X-ray of an AI business in nine blocks across three levels: strategy (workflow wedge, distribution, retention), economics (inference economics, heavy users, unit economics, burn), and synthesis (stress test, survival verdict). The Canvas is the diagnosis: where the leak is. The Codex is the discipline: what you do about it without lying to yourself.
Together they answer the one question that decides the fate of an AI business: “Will my company survive 10× growth — and why exactly?” Not in a hundred metrics. In one causal sentence the founder, the team, and the investor can quote back. The full AI Survival Canvas walkthrough is the companion article in this series.
Survivability as advantage, not just survival
One last turn. Up to here, the Codex has been about not dying. But in volatile AI, survivability stops being purely defensive and becomes offensive. Most of your competitors will break before you do — at the next vendor price update, the next model refresh, the next PMF shift, the next reasoning leap that turns last year’s edge into this year’s baseline. The founder who is ready doesn’t just survive. They inherit the market the others vacated.
That is the real argument for the Codex. It is not protection from defeat. It is a competitive advantage in an era when lasting is harder than launching.
What to do next
Reading isn’t doing. Three options, in ascending order of investment: