Max Hodak: How Startups Build Speed

Why does it matter? Because the world's best scientists are watching their companies die — and the cause is almost never the science

The CEO of a company that makes retinal implants — chips that let blind patients finish 300-page novels — wants to talk about procurement software. That's not a detour. Max Hodak spent two decades at the edge of neuroscience and came out convinced that deep tech startups almost never fail because the technology doesn't work. They die because nobody built a good purchasing system, a coherent budget structure, or a recruiting process that scales.

• A team compounding knowledge weekly will permanently outpace a monthly-learning competitor — the math forecloses any comeback
• Approving individual purchases is the wrong place to exercise cost control; the delay already costs more than any savings
• A PageRank-style continuous peer vote called eigenreviews produces cleaner performance signal than any annual review cycle
• Vibe coding your own procurement system is no longer irrational — it may now be the highest-leverage early investment in speed

Deep tech companies almost never die from bad science — they die because nobody built a procurement system

Stellar scientists and engineers die on the vine all the time. Not because their technology failed — because their organizations couldn't execute. Hodak is unequivocal: "It is uncommon that deep tech companies fail because the technology doesn't work. They fail because once you end up with this organization of hundreds of people and hundreds of thousands of square feet of physical infrastructure, you haven't built the systems to manage that."

Science — Hodak's company — has a reputation for unusual speed in a sector not known for it. The product is a subretinal prosthesis that restores vision in patients whose rods and cones have failed; it finished major clinical trials last year, landed on the cover of Time, and helped a patient through a 300-page novel. Ask Hodak how the company moves so quickly and you get an answer that has nothing to do with talent or scientific advantage.

"How does that happen is mostly not that we are smarter — it is infrastructure like this. That is how speed is built."

The infrastructure he means is unglamorous: purchasing systems, budgeting structures, recruiting pipelines, performance review processes. Founders who treat these as administrative overhead — something to handle after the science is right — are building on sand. The companies that survive are the ones that figure out early that operational systems are engineering problems, not support functions. Build them badly and you end up with hundreds of millions in funding, hundreds of people, and no ability to connect strategy to execution.

A weekly-learning team will permanently outpace a monthly-learning competitor — the gap only widens

The math is decisive. "If you can learn one thing every week, and there's a competitor that's learning a thing every month, they will never matter." Not eventually — never. A 4x iteration advantage closes no gaps. It widens them indefinitely.

This shapes how Hodak evaluates competing approaches. When two technical paths diverge and one supports faster experiment cycles, he argues you should weight that heavily — "even if the other approach has significant redeeming characteristics, you should really consider going with the shorter iteration cycle because the compounding effect is just so dramatic."

The practical consequence is cost attribution. For years, Science ran biological experiments in a foundry without knowing what each iteration cost. When nobody tracks it, "experiments are free — it doesn't cost dollars, it costs media, and media comes from the fridge." Once their internal platform Helix built full cost attribution through the process, the real number surfaced: $40,000 per wafer iteration. Suddenly, experiment choices became economic decisions. That's the mechanism — not working harder or hiring smarter, but making the cost of each loop visible so teams can make rational trade-offs about which loops to run and which to kill.

Approving the $3,000 power supply is already too late — the delay cost more than the purchase ever could

Your engineer needs a power supply. It costs $3,000. You see the message and think: there's an auction in three days, maybe it comes in at $1,500. You wait.

"If you wait a week to get a power supply half off, you have certainly dwarfed any possible benefit from getting it."

The purchase price is almost incidental. A week of blocked progress from a highly-paid researcher costs multiples of any savings — plus the cultural signal that tools aren't readily available, plus the implicit comparison to any well-run company where a $3,000 equipment request never reaches a founder's inbox. "Spending review has to come earlier. You have to have some concept of budgeting."

Moving the control layer upstream means giving teams a budget — a defined bucket they own — and letting them make trade-offs inside it. The founder's job is to design the budget, not approve the line items. Helix gives Science teams cost visibility at the experiment level, so researchers make economically rational choices because they have economic information. That's the actual mechanism of burn control: not approval chains, but visibility distributed to the people who can act on it. When nobody knows what things cost, everything feels free. That's a culture problem that compounds silently.

Any small group placed in front of your hiring funnel will eventually strangle the whole organization

When a candidate applies to Science, seven or eight current employees get pinged automatically — people whose backgrounds resemble the applicant's. Not the recruiting team. The whole company votes.

"When a person applies, the system picks out seven or eight current employees that it thinks look something like their backgrounds, and it pings them all for votes."

The design rationale is twofold. First, throughput: top-of-funnel review scales with application volume, and any centralized filter caps the organization. "If you place any small group of employees or any one person in the way as a bottleneck on this, they will absolutely bottleneck the whole rest of the organization." Second, judgment averaging: individuals have biases, hiring intuitions vary enormously, and "you want ways to average over the judgment of the rest of your team."

Science's four-step process: company-wide vote → phone screen (drawn from a company-wide pool, evaluating judgment, horsepower, and agency rather than team fit) → homework (AI-resistant where possible, with a high ceiling and naturally scorable outputs) → full interview. Seventeen percent of applicants reach the phone screen. On-site-to-offer conversion should stay above 25% — below that, interviewers are burning time on candidates who shouldn't have made it that far. The commercial ATS couldn't route applicants to employees with similar backgrounds. So they built Helix to do it.

Annual performance reviews surface nothing you didn't already know — a PageRank algorithm run every six weeks does it better

Conventional reviews confirm what you already knew and traumatize the organization in the process. "Based on my experiences, this is a very disruptive process that doesn't tend to surface issues that you don't already know about but haven't acted on."

Eigenreviews work differently. Every four to six weeks, employees receive a single question through Helix: knowing how this person turned out, would you vote again for their hire today? It's the same question as the initial hiring vote, which makes the feedback loop coherent across the full employee lifecycle. Scores are then weighted by a graph algorithm: your vote carries more weight if the people you've rated positively are themselves highly rated. "Your vote should be weighted more highly if everybody else has rated you highly. And the astute may notice that this looks a lot like the original Google algorithm, PageRank — an idea called eigenvector centrality."

To prevent voting cliques from gaming the signal, the system runs a thousand Monte Carlo iterations with random edge dropout. Two peaks in the score distribution flag a bloc worth investigating. The result is a continuous, algorithmically weighted performance signal updated with roughly a month lag — distributed across the whole company, without the semi-annual organizational disruption. "I've become convinced that this is more or less the right way to do performance reviews."

Cheating off your neighbor's test drags your grade toward average — and average is never enough to win

Startup success lives in the long tail. Reaching it requires being differentiatedly right about something, which means not averaging toward consensus. "In school, if you cheat on the test by looking over at your neighbor, your grade will be dragged towards the average of the class. That is not good enough to succeed in startups."

Generic startup advice is, by construction, averaged wisdom. At the moments it matters most, it pulls you toward median outcomes. Hodak describes a specific feeling that arrives years in: hundreds of millions on the line, a high-stakes decision, and no one to ask. "You'll get to a key point years in and you'll look for advice and there's nobody to ask."

That moment will come. The preparation isn't consuming more advice — it's spending years calibrating your judgment with real stakes attached. Working alongside someone with empirically good judgment, as Hodak credits his time at Neuralink, trains your filters in a way that reading post-mortems cannot. Use outside counsel to sharpen those filters. But when the call comes that only you can make: "You cannot delegate your judgment. As the CEO, you must always make decisions that make sense to you, no matter how much momentum or inertia the alternatives seem to have."

Vibe coding a procurement system is no longer irrational — it might be the highest-leverage investment a seed-stage company makes

SpaceX and Tesla run manufacturing and R&D through an internal platform called Warp Speed. YC has internal software Hodak credits as central to how it functions at scale. The pattern: companies that built operational software around how they actually think, rather than adapting processes to whatever commercial SaaS was available.

"The fact that you can vibe code this now makes it a reasonable thing to think about. Historically, software has been so expensive, you would have had to buy it."

Commercial tools impose their logic on your processes. An off-the-shelf ATS can't route applicants to employees with similar backgrounds. An ERP system won't cost-attribute your cell-line experiments. When AI coding tools collapse the cost of building custom software, the build-vs-buy calculus shifts permanently for operational systems. Hodak acknowledges that telling investors at the seed stage you're vibe-coding a procurement platform would raise serious board questions. His answer: control the company and build what gives you speed. "Placing that with software, we were able to explore voting mechanisms and fairly detailed voting mechanisms that can make smart inferences about who would know about an applicant — things that you can't really do with the commercial software."

When you're stuck in a local minimum, planning won't get you out — injecting arbitrary action will

"Action produces information. This idea is, I think, much deeper than it sounds."

In physics, action is a specific quantity, and Hodak means this almost literally: push something into the world, and you generate information that didn't exist before. When a company is genuinely stuck — recycling the same options, analysis going nowhere — what it needs is state change. He's watched companies where removing one person, individually strong but wrong for what the organization needed at that moment, "unblocks the company and allows it to enter a new phase." The removal worked not because it was obviously correct, but because it created information and options that deliberation couldn't produce.

The action space is almost always larger than it appears from inside a local minimum. The data needed to pick the next move often only exists after you generate it by moving.

The operating system of a company is as strategic as the product itself

What eigenreviews, company-wide hiring votes, and custom procurement software share is a conviction that how a company buys, hires, reviews, and learns is not peripheral to building something great — it is the substance of it. The founders who treat operational infrastructure as a primary engineering problem from the earliest days won't just move faster. They'll build organizational compounding that competitors on slower loops genuinely cannot close. Speed isn't the outcome. It's the moat.

---

Topics: startups, deep tech, operations, hiring, performance management, procurement, brain-computer interfaces, organizational design, iteration speed, internal tooling, YC Startup School