The cofounder is the most expensive line on your cap table

I price private companies for a living. I have run the numbers on more than a thousand of them, and worked on M&A deals up to roughly $26 billion. When I open a model, the revenue, the margins, and the comparable multiples are the easy part. The hardest input, and the one most founders treat as an afterthought, is the founding team. It is the largest line I cannot model cleanly, and it moves the answer more than almost anything else on the page.

Here is the part founders miss. Your cofounder decision is not a relationship decision that happens to have business consequences. It is a valuation decision that happens to feel like a relationship. Investors price the consequences of it, whether you negotiated them carefully or shook hands in an afternoon. So before you talk about who you like, talk about what it costs to get this wrong. Because the wrong cofounder, locked in at the wrong terms, is the single most expensive line on a cap table that has not raised a dollar yet.

Why this is a valuation decision, not a friendship

Start with the cleanest academic finding I know. In a study of 1,476 founders across 511 ventures, Hellmann and Wasserman found that roughly one third of teams split their equity equally, and that equal splitting is associated with lower pre-money valuations in the first financing round. The value at stake from a default-equal split is about 10 percent of firm equity, 25 percent of the average founder's stake, or roughly $450,000 in net present value. The penalty is worse for teams that decide fast, in under a day, the so-called quick handshake. Read that again. A sloppy equity decision is literally priced into your company by the people writing the check.

It compounds from there. CB Insights' post-mortem analysis ranks "not the right team" among the top reasons startups die, and lists cofounder and investor disharmony as its own failure line. You have probably heard the "65 percent of startups fail from cofounder conflict" stat. Be honest about it. That number traces back to a 1989 survey of venture capitalists about troubled portfolio companies, where about 64 percent put management-team problems in their top three reasons for failure. Still a brutal signal. It just measures VC perception of team breakdown broadly, not cofounder conflict specifically. The direction is not in dispute. Teams are where the money dies.

And founders lose control faster than they plan for. Wasserman's analysis of 212 American startups found that by year three, half of founder-CEOs were no longer the CEO, and fewer than a quarter still led at IPO. The large majority were pushed out, not leaving by choice. The structure you set on day one decides who survives. That is why I read a founding team the way I read a cap table. Slowly, and looking for what breaks.

What to actually look for

Forget "find someone you trust." The data says trust by itself is a trap. In Wasserman's sample of nearly 10,000 founders, each friendship in a founding team raised the rate of founder turnover by 28.6 percent. Teams of former coworkers were the most stable. Teams of relative strangers held together better than teams of friends. The mechanism is simple and human: friends and family avoid the hard conversations to protect the relationship, so the disagreements that should surface in month one surface in year two, when there is a cap table to fight over.

So work together before you marry. Not coffee. A scoped, paid, month-long trial project with an explicit go or no-go decision at the end. YC pushes exactly this, and warns against picking "some random dude you barely know." A real trial surfaces work ethic, how someone handles being wrong, and whether your risk tolerance matches, before any of it is welded to your equity. Use a structured set of hard questions up front. How much money do we each need to take home. What happens if one of us wants to sell in year three. Who decides when we disagree. If a conversation that direct ends the partnership, it was going to end anyway, and you just saved yourself $450,000 and two years.

On skills, resist the reflexive "complementary skills" advice. It is overrated. Two strong technical founders can be a great team. What you actually need is at least one person who can build the thing. If you are non-technical, that is non-negotiable. What you want aligned is values, work ethic, and appetite for risk. What you want different is the part of the job you each hate.

AI-era team composition and the roles that matter

The headcount math has changed, and it changed the team design problem with it. AI-native startups run dramatically leaner than the SaaS playbook assumed. Reporting on the new lean-AI companies has found an average of just 19 employees, a typical founding team of two, two to three engineers hired in year one, and roughly five people added per year after that. Seventy-four percent were profitable. Revenue per employee ran $1.2 to $1.6 million, against about $200,000 for traditional SaaS. Cursor crossed $100 million in ARR within about 20 months at 40 to 60 people. Midjourney reached over $200 million annualized with around 40 employees and zero venture money. Small teams are now capable of outcomes that used to require a hundred people.

That means you are not staffing a company. You are choosing the two or three roles that have to live inside the founding team versus the ones you can contract. The roles that genuinely matter early: someone who owns the applied-AI product, meaning the person who turns models into something a customer pays for, not the person who trains them from scratch. A domain expert, because in an AI company the edge is rarely the model and almost always the proprietary problem understanding and data. And someone who owns go-to-market, because lean does not mean nobody sells.

Here is the contrarian piece. You probably do not need a research-grade ML cofounder. Frontier models are a commodity you rent through an API. What you need is product and domain judgment about where those models create defensible value. If you do require deep custom modeling, you can often contract it before you give away founder equity for it. But understand the trade you are making. A tiny team is also a concentrated key-person risk. When two people generate eight figures of revenue, the departure of one is not an HR event, it is a valuation event. I discount for it explicitly. Build redundancy into knowledge and relationships early, or the leanness that makes you efficient also makes you fragile.

Equity and decision rights, read like an investor

Now the part everyone gets wrong on day one. The NBER data penalizes equal-and-fast splits. YC, correctly, argues for near-equal splits because the company's value is built by the seven to ten years of work ahead of you, not the head start one of you had in month one. Michael Seibel's line is right: small variations in year one do not justify wildly different founder splits, and refusing to split fairly can be a signal you picked the wrong partner. These are not in conflict. The resolution is equal but deliberately negotiated, never equal by reflex. Have the real conversation, then land where you land.

Then protect it with structure, because the split is not the risk, the lack of mechanism is. Vesting, four years with a one-year cliff. Twenty-five percent at year one, then monthly. This is the most important sentence in this essay. It means a cofounder who leaves in month eight walks away with nothing, instead of holding 30 percent of your company as dead equity that poisons every future round. Investors read dead equity on a cap table as a reason to pass or to mark you down, and they are right to. Decide who is the single decider when you deadlock, and write it down. Two equal owners with no tiebreaker is not partnership, it is a pending lawsuit. None of this is about distrust. It is about making the team an asset on the cap table instead of a liability priced against you.

Finding a cofounder with no inherited network

I built three companies in the United States as an immigrant from Krakow, with no warm intros and no insider circle to draw a cofounder from. So I will tell you plainly: the "find someone you already trust" advice quietly assumes you inherited a network full of qualified people. Most founders did not. I did not.

What works without a network is the same thing the data already told us. Manufacture the shared work history you do not have. Go where builders already gather and ship something small with people, an open-source contribution, a hackathon, a consulting project, a paid trial. You are not looking for a friend. You are looking for a former coworker, and if you do not have one, you create the coworker relationship on purpose before you create the company. It is slower. It is also the version that does not blow up in year two, which is the only version worth having. The absence of a network is not the disadvantage it looks like. It forces you to select on evidence instead of on familiarity, and evidence is exactly what an investor pricing your team is looking for too.

So choose the way someone who has to price the consequences would. The cofounder you pick, and the terms you pick them on, will sit on your cap table longer than any product decision you make this year. Treat it with at least that much seriousness, and you have removed the most expensive mistake available to you before you have written a line of code.

Common questions

Should cofounders split equity 50/50?

Near-equal is often right, because most of a company's value comes from the seven to ten years of work ahead, not the head start one of you had. The danger is splitting equally by reflex in an afternoon. NBER data ties equal-and-fast splits to lower first-round valuations, roughly $450,000 in lost value. Have the hard conversation, decide deliberately, and always protect the split with four-year vesting and a one-year cliff.

Do I need a technical or ML cofounder for an AI startup?

You need someone who can build the product. You usually do not need a research-grade ML cofounder. Frontier models are a commodity you rent through an API, so the scarce skill is applied-AI product judgment and domain understanding about where models create defensible value. If you need deep custom modeling, you can often contract it before giving away founder equity. If you are non-technical, a builder in the founding team is non-negotiable.

Is it a bad idea to start a company with a friend?

The data is unkind here. In Wasserman's sample of nearly 10,000 founders, each friendship in a founding team raised founder turnover by 28.6 percent, and friend teams were the least stable. Former coworkers were the most stable. The reason is that friends avoid the hard conversations to protect the relationship, so conflicts surface late, when there is a cap table to fight over. Work together professionally first, then decide.

How big should an AI startup's founding team be?

Smaller than the old SaaS playbook assumed. Reporting on the new lean-AI companies has found an average of 19 employees, a typical founding team of two, and two to three engineers hired in year one. Cursor crossed $100 million ARR at 40 to 60 people. Midjourney passed $200 million annualized with around 40. Choose the two or three roles that must live in the team, and contract the rest.

Why does vesting matter so much for cofounders?

Because the split is not the real risk, the lack of a mechanism is. Standard vesting is four years with a one-year cliff: 25 percent at year one, then monthly. It means a cofounder who leaves in month eight walks away with nothing, instead of holding a large dead-equity stake that poisons every future round. Investors read dead equity on a cap table as a reason to pass or mark you down, and they are right to.

Related reading

• Building without warm intros: why the outsider's rigor became my edge.
• What I learned about M&A pricing after running 1,000+ valuations.
• I left billion-dollar deals to price ten-million-dollar businesses.

Go deeper

If you are setting up a founding team, or staring at a cap table you are not sure you priced right, let's work the actual numbers together, or read more of how I think about it.

Tomasz Felpel is an investor, founder, and advisor in private markets and healthcare, based in New York. He is a three-time founder of Value Alpha, an AI-powered private-markets valuation platform, Sonnerie VC, an early-stage healthcare venture firm, and Pond. Previously he led corporate development and M&A at Fortune 500 scale, pricing more than 1,000 private companies. Columbia Business School EMBA. Read the full story.