Fifteen axioms we keep returning to while building a company, statements we treat as ground truth because every hard decision eventually derives from them. Covers survival, revenue as the only metric, failing fast, co-founder trust, timing, team composition, curiosity, excellence, and the three-month feedback loop.
Every framework we tried while building a company eventually broke, except a small set of statements that survived contact with reality every single time. We stopped calling them lessons and started calling them axioms: things we no longer argue about, we just derive from them.
In mathematics, an axiom is a statement you accept without proof because everything else is built on top of it. You do not re-derive the axioms every morning. You use them.
Building a startup floods you with decisions that have no clean answer: take this client or not, hire this person or not, kill this feature or push one more month. If you reason about each decision from scratch, you drown. What saved us was compressing our experience into a small set of ground truths and deriving decisions from them instead. When a decision contradicts an axiom, the decision is wrong, not the axiom; and on the rare occasion an axiom actually breaks in the arena, that is the most valuable information you will get all year, because you just found out something load-bearing was false.
This post is that set. Fifteen axioms, in three groups:
None of these are provable. All of them are falsifiable, and so far the arena has not falsified any of them for us.
The first five axioms are the ones we would tattoo somewhere visible. They define what a business is before any strategy discussion makes sense.
Every dramatic startup death story hides a mundane final frame: a founder deciding not to show up anymore. Markets crash, clients churn, co-founders leave, runway burns. None of these events kill a business directly. They drain resources and morale, and then a human converts that drain into a decision.
We can state this as a survival process. Let $\text{alive}_t \in {0, 1}$ be whether the business exists at time $t$, and let $\text{quit}_t \in {0, 1}$ be the founder’s decision that period:
\[\text{alive}_{t+1} = \text{alive}_t \cdot (1 - \text{quit}_t)\]External shocks appear nowhere in the transition. They influence how tempting $\text{quit}_t = 1$ becomes, but the variable itself has exactly one owner: you. Death is an absorbing state that can only be entered voluntarily.
The operational consequence is boring and absolute: always show up. Showing up daily is how you hold $\text{quit}_t$ at zero through the periods when every signal says stop. Most competitors are not out-executed; they are outlasted.
Users, followers, press coverage, awards, incubator selections, pitch competition wins: all of these feel like progress and none of them are the metric. A business has exactly one objective function:
\[J = R = \sum_{c \in \text{customers}} p_c \cdot q_c\]where $p_c$ is the price customer $c$ pays and $q_c$ is the quantity they buy. Every other number you track is worth exactly its marginal effect on this sum, $\partial R / \partial m$ for metric $m$. If that derivative is zero, the metric is decoration.
This axiom sounds obvious and is violated constantly, especially in ecosystems where grants, competitions, and demo days pay out before customers do. You can spend two years optimizing everything except $R$ and feel productive the entire time. Revenue is the only signal that cannot be gamed by your own wishful thinking, because it requires a stranger to part with money.
Your idea is a hypothesis, not a child. The moment you love it, you start defending it from the market instead of testing it against the market, and defense is expensive.
The math favors iteration speed over idea quality. Suppose each attempt has probability $p$ of finding something that works, and one iteration takes time $\tau$. Over a horizon $T$ you get $T/\tau$ shots, and the probability at least one lands is:
\[P(\text{hit}) = 1 - (1 - p)^{T/\tau}\]The exponent is where the leverage lives. Doubling $p$ requires being twice as smart as the market, which you are not. Halving $\tau$ requires killing bad ideas faster, which is entirely under your control. Loving your idea inflates $\tau$: you run one more survey, add one more feature, wait one more quarter, all to postpone a verdict you already suspect.
Fail fast does not mean be careless. It means route every idea to the cheapest experiment that can falsify it, accept the verdict, and reload.
In an established industry you can bet on the plan, because the map is known. In an innovation business there is no map: the product will pivot, the market will surprise you, the technology will shift under your feet. The plan you pitched is guaranteed to be wrong in the specifics. The only thing that persists through every pivot is the team.
Formally, the expected value of an innovation venture conditions almost entirely on the people:
\[E[V \mid \text{team}, \text{idea}] \approx E[V \mid \text{team}]\]Conditioning on the idea adds little information, because the idea will be replaced two or three times before anything works. This is why experienced investors say they bet on the jockey, not the horse, and why acqui-hires exist: even when the product fails completely, the team retains market value.
The uncomfortable inverse also holds. A brilliant idea attached to a team that cannot adapt is worth roughly the team, which is to say, not much.
There is an Indonesian phrase for the bar we use: a co-founder should be someone yang lu siap ajak keliling rumah dan intip kamar, someone you would let walk around your whole house and peek into your bedroom. Not the living room you clean for guests. The bedroom. The mess, the finances, the fears, the ugly drafts.
The economics of this are direct. Every unit of missing trust becomes monitoring overhead. Let $T \in [0, 1]$ be trust between partners and decompose the total cost of running the company:
\[C_{\text{total}} = C_{\text{work}} + (1 - T) \cdot C_{\text{monitor}}\]At $T = 1$ the second term vanishes: decisions are delegated in one sentence, money moves without triple-checking, bad news travels instantly because nobody fears blame. As $T$ drops, you pay a tax on every single interaction: contracts get longer, updates get filtered, and eventually the founders spend more energy watching each other than watching the market. We wrote about this same mechanism at the ecosystem scale in our previous post; inside a founding team the effect is faster and more lethal, because a startup has no institutional buffer to absorb the friction.
Skills can be hired. Trust cannot. Choose the partner first, at the bedroom-tour standard, and build everything else on top.
The second five axioms deal with time: when to start, how to learn, and who to learn beside. The common thread is that motion precedes knowledge, not the other way around.
Everyone waits for the moment when the savings are sufficient, the market is ready, the product is polished, the timing is right. Model your perceived readiness as a saturating curve:
\[\text{readiness}(t) = 1 - e^{-\lambda t}\]It increases forever and reaches certainty never. So if your launch bar is $\theta \geq 1$, if what you are waiting for is knowing it will work, the waiting condition $\text{readiness}(t) \geq \theta$ is satisfied at no finite $t$. You will study the market for one more quarter indefinitely, and each quarter the opportunity cost accumulates linearly while your readiness gains decay exponentially.
The people who look like they timed it perfectly are, on inspection, people who started at an ordinary moment and were still standing when the moment arrived. You cannot schedule luck. You can only maximize your exposure to it, and exposure requires being in the arena before you feel ready.
At launch, several of your assumptions are wrong and you do not know which. Pricing, channel, segment, feature priority: some subset is unfit, guaranteed. This is not a flaw in your preparation; it is a property of the problem. The loss function you are minimizing is only observable in production.
The honest model of building a company is online learning, not offline planning:
\[\theta_{t+1} = \theta_t - \eta \, \nabla L_{\text{market}}(\theta_t)\]where $\theta_t$ is your current configuration of the business, $\eta$ is how aggressively you adapt, and $L_{\text{market}}$ is a loss you can only query by acting: shipping the price, running the channel, calling the customer. No amount of analysis evaluates $L_{\text{market}}$ from your desk, because your desk samples from the wrong distribution: the imagined market instead of the real one.
The axiom cuts both ways. It forgives the broken variables you launched with, and it condemns the founder who refuses to update $\theta$ after the market has returned its gradient.
Three brilliant backend engineers founding a company together are, functionally, one backend engineer with more opinions. Team capability behaves like a coverage function: submodular, with sharply diminishing returns for overlap. For a team $S$ and a candidate member with skills $s$:
\[f(S \cup \{s\}) - f(S) \longrightarrow 0 \quad \text{as } s \text{ overlaps } S\]The marginal value of a new member is not their absolute strength; it is the area they cover that nobody else does. A company needs someone who builds, someone who sells, someone who operates, and it needs them from day one, because in the early days every one of those surfaces is on fire simultaneously.
Duplicate skillsets carry a second, quieter cost: identical people compete for the same decisions. Two visionary product minds will fight over the roadmap; a builder and a seller almost never fight, because their territories barely intersect. Complementarity is not only about coverage, it is about peace.
A great mentor compresses your mistakes: they have seen your situation before and can hand you the answer for free. If you can get one, take it. But treat mentorship as an accelerator term, never a dependency. Your knowledge grows as:
\[\frac{dK}{dt} = r_{\text{read}} + r_{\text{do}} + m \cdot r_{\text{mentor}}, \qquad m \in \{0, 1\}\]The first two terms are always available. Books are mentorship at one-thousandth the price: the best founders and operators of the last century wrote down exactly what they learned, and the material is sitting there mostly unread. Doing is the other half; Axiom 7 already established that the market tutors anyone who ships.
The failure mode this axiom guards against is the founder who postpones building until they find the right advisor, incubator, or program. That is Axiom 6 wearing a different mask. With $m = 0$ the derivative is still positive. Read stuff. Ship stuff. The curve compounds either way.
Outlier outcomes live in the tail of the distribution, and you do not reach the tail by exploiting known strategies; everyone exploits the known strategies, which is precisely why their returns are average. Reaching the tail requires sustained exploration, and the engine of sustained exploration is curiosity.
The bandit literature gives the clean formalization. An upper-confidence-bound agent picks actions by:
\[a_t = \arg\max_a \left( \mu_a + c \sqrt{\frac{\ln t}{n_a}} \right)\]where $\mu_a$ is the known payoff of action $a$, $n_a$ is how often it has been tried, and $c$ is the exploration coefficient: the weight placed on the unknown. Average operators run small $c$: they do what worked last quarter. Outliers run large $c$: they read outside their field, take the odd meeting, prototype the tangent, and therefore occasionally find the arm nobody has pulled.
The phrasing of this axiom is deliberate: converge toward extreme curiosity. Most people are not born with large $c$, and that is fine; it is a trainable parameter, raised by deliberately scheduling exploration until the habit becomes appetite. If you intend to be an outlier, this convergence is not optional.
The final five axioms decide whether everything above compounds or dissipates. They are about standards, incentives, and the strange bookkeeping by which effort returns to you.
On your worst days, talent does not show up, but a system does. A checklist, a weekly cadence, a standard operating procedure: these guarantee a floor on output that is completely independent of your mood, health, or motivation that morning:
\[E[\text{output} \mid \text{system}] \geq \text{floor} > E[\text{output} \mid \text{mood}]\]This axiom is not an insult; it is permission. You might start mediocre, most people do, and the correct response is to stick to a system, because the system keeps you shipping through the long stretch where your skill has not caught up with your ambition. The mediocre operator with a system outlasts the talented operator without one, which, by Axiom 1, means the mediocre operator wins.
The system is the starting point, not the ceiling. Which is exactly where the next axiom takes over.
Big opportunities do not evaluate your pitch, they evaluate your track record, and your track record is built entirely out of small deliveries. Reputation compounds multiplicatively, not additively. Over $n$ pieces of work with quality $q_i \in [0, 1]$:
\[R = \prod_{i=1}^{n} q_i\]A product, not a sum. Nine excellent deliveries and one sloppy one do not average to nearly-excellent; the sloppy one multiplies through and drags the whole product down, because the client who received it tells the story loudest. The big doors are guarded by people who inspect exactly the smallest thing you ever shipped: the follow-up email, the invoice formatting, the demo that had to work at 8 a.m.
So the key to the big door must be the small one: treat the tiny engagement with the same standard you would give the dream contract. There is no separate audition for the big stage. The small work is the audition, running continuously, whether you notice it or not.
Every person who touches your company, co-founder, employee, investor, client, partner, arrives with their own objective function, and it is never identical to yours. The investor wants a return on a fund timeline. The employee wants growth and rent money. The client wants their problem gone. None of them wake up wanting your company to win; they want their objective met, and your company is one of the instruments.
A collaboration survives only while it satisfies every participant’s individual rationality constraint:
\[u_i(\text{deal}) \geq u_i(\text{outside option}) \qquad \forall i\]The moment any party’s payoff drops below their outside option, the deal starts dying, whatever the contract says: effort quietly drops, priorities drift, and the exit gets negotiated by behavior long before it is negotiated in writing.
The operational rule: surface the objectives first, before the work starts. Ask directly what each party needs from the arrangement, design so every constraint holds, and revisit when circumstances shift. Alignment is not a vibe; it is a system of inequalities, and you want all of them satisfied on day one.
Founders agonize over picking the right project and barely think about picking the right room. The weighting is backwards. What you work on sets this year’s output; where and with whom you work set the growth rate of every year after. Model your trajectory as compound growth:
\[K(t) = K_0 \cdot (1 + r_{\text{env}} + r_{\text{people}})^{t}\]The project determines the base $K_0$, a one-time constant. The environment and the people determine the exponent. A mediocre project in a room full of ambitious, curious, trustworthy people compounds into remarkable outcomes: the room upgrades your standards (Axiom 12), your curiosity (Axiom 10), and your deal flow, automatically, every day. A brilliant project in a stagnant room decays toward the room.
This is also the axiom that quietly contains three of the others: the co-founder bar of Axiom 5, the complementarity of Axiom 8, and the alignment of Axiom 13 are all instances of one rule. The people are the position. Choose the room before the task.
Business runs on a lag. The client you treat well today refers someone in a quarter. The corner you cut today becomes the outage, the churned account, or the awkward renegotiation in a quarter. Almost nothing you do in a startup pays or punishes immediately; the ledger settles later, and empirically the settlement window is about ninety days:
\[\text{outcome}(t) = f\big(\text{effort}(t - 90)\big)\]Today’s results were purchased three months ago. Today’s effort is purchasing a day you cannot see yet.
This axiom changes how you read your own dashboard. A great month is not evidence that you are working well now; a dead month is not evidence that you are working badly now. Both are echoes. It also closes the loop back to Axiom 1: the founder who quits in a quiet stretch is usually abandoning a harvest that was already planted, ninety days from surfacing. Keep showing up. The ledger is slower than you, but it does not lose entries.
Fifteen axioms, three groups, zero proofs. We hold them the way mathematicians hold theirs: not because they are provable, but because everything we have managed to build derives from them, and because the arena, which falsifies almost everything, has not falsified these. Steal the list, test it against your own arena, and replace any entry the moment reality breaks it. That replacement rule is the unwritten sixteenth axiom.