AVERAGE IS NOT GOOD ENOUGH

Max Hodak’s core argument:

Startups rarely fail because founders lack ideas. They fail when the organization cannot turn judgment into execution fast enough.

Speed is built. Infrastructure is the operating system behind it.

THE CORE THESIS

1. Speed determines success or failure.

2. Speed is determined by infrastructure.

3. Infrastructure means purchasing, budgeting, recruiting, reviews, quality, software, and decision systems.

The boring systems often decide whether the ambitious vision survives.

INFRASTRUCTURE

PROFESSIONALS TALK LOGISTICS

Picasso’s lesson: critics discuss form and meaning; artists discuss where to buy cheap turpentine.

For startups, the equivalent is simple:

Ideas matter. But execution depends on logistics—the systems that let talented people move without waiting.

BUYING THINGS BECOMES A SYSTEM

A founder can buy something with a credit card.

But what happens when the 17th employee needs a $3,000 power supply?

If every purchase needs founder approval, the founder becomes the bottleneck. If nobody controls spending, burn becomes invisible.

BUDGET EARLIER, NOT PURCHASE-BY-PURCHASE

The wrong place to control spending is every individual purchase.

Give teams a clear budget and let them make trade-offs inside it.

The goal is not to debate $1,500 vs $3,000 tools. The goal is to define the resource envelope before the decision happens.

PROCUREMENT IS A SPEED ENGINE

B2B purchasing can require vendor setup, insurance, certifications, quotes, purchase orders, invoices, and follow-ups.

Without ownership, a simple tool can take weeks to arrive.

Fast companies actively manage procurement lead time as an operating metric.

KNOW WHAT WORK ACTUALLY COSTS

Deep-tech teams buy materials in bulk, then split them across experiments.

If nobody knows what one experiment costs, the experiment feels free.

Cost attribution lets founders understand pricing, runway, unit economics, and which iteration loops are consuming the company.

BUILD THE COMPANY OPERATING SYSTEM

Science built internal software called Helix.

Purchasing, manufacturing steps, batch records, quality, and other workflows live in one system.

That creates traceability: what happened, what it cost, who did it, and how each step connects to the final product.

BIG FUNDING CAN HIDE SMALL LEAKS

A large round can make thousands of dollars feel insignificant.

But payroll, space, tools, materials, and repeated experiments compound quickly.

A startup can raise tens of millions and still discover that its actual research budget is far smaller than the headline number suggests.

HIRING

START WITH THE SCENE

Strong startups often emerge from a scene: a lab, company, community, technical movement, or group of people already speaking the same language.

Your first hires often come from that network.

They already understand the context and can move with less translation.

HIRING NEEDS A DEFINED PROCESS

There is no single perfect hiring process.

But having no consistent process is a clear mistake.

Applications, reviews, interviews, and decisions quickly consume founder and team time. A repeatable system protects attention while improving signal quality.

SCIENCE USES 4 HIRING STAGES

1. Initial companywide voting

2. Phone screen

3. Homework or practical test

4. Full interview

The goal is to collect enough signal to make a strong decision while preventing a small recruiting team from becoming the company bottleneck.

DISTRIBUTE THE FIRST REVIEW

When someone applies, Science routes the candidate to several employees with relevant backgrounds.

They vote on whether the candidate should advance.

This spreads judgment across the company and keeps the top of the hiring funnel moving within roughly 24–48 hours.

SCREEN FOR 3 TRAITS

Science’s phone screen looks for:

1. Judgment — do you make good decisions in ambiguity?

2. Horsepower — can you learn and solve hard problems?

3. Agency — can you cause the world to look more like the outcome you want?

DESIGN TESTS WITH A HIGH CEILING

The best homework tasks do not saturate easily.

They produce outputs that can be scored on a few clear dimensions, making exceptional performance obvious.

AI use is not automatically a problem. A strong task should still reveal who can push beyond the normal frontier.

PROTECT THE ON-SITE INTERVIEW

By the full interview stage, the candidate pool should already be strong.

If almost nobody converts to an offer, earlier filters are weak and the team is wasting expensive time.

The funnel should become increasingly selective before the highest-cost stage.

PERFORMANCE

MARKET FEEDBACK IS TOO SLOW

Eventually, the market tells you whether your hiring worked.

But that feedback can take years.

Management needs earlier signals—synthetic gradients—that show whether the organization is improving, where problems are forming, and which decisions need correction.

ASK ONE POWERFUL QUESTION

Every few weeks, employees at Science are asked a simple question about people they work with:

Knowing how this person turned out, would you vote again today for their hire?

The repeated answer creates a continuous signal instead of waiting for an annual review cycle.

USE THE NETWORK, NOT ONE MANAGER

Science builds a graph from peer feedback and weights judgments based on the credibility of the reviewers.

The approach is inspired by network-centrality ideas like PageRank.

The goal is not one manager’s opinion. It is a distributed, continuously updating view of organizational performance.

ITERATION SPEED

WEEKLY LEARNING BEATS MONTHLY LEARNING

If one startup learns something important every week and a competitor learns once a month, the slower company can become irrelevant.

Iteration frequency compounds.

When two approaches are viable, strongly consider the one that produces reliable feedback much sooner.

SPEED IS BUILT UNDERNEATH THE PRODUCT

Fast execution is rarely just smarter people working harder.

It comes from systems that remove waiting:

• Faster purchasing
• Clear budgets
• Better recruiting
• Reliable information
• Strong internal tooling
• Short feedback loops

DEEP-TECH OFTEN FAILS ORGANIZATIONALLY

Hodak argues that deep-tech companies often do not die because the core technology is impossible.

They die when hundreds of people, large facilities, complex quality systems, and huge spending become impossible to coordinate.

The founder’s job is to organize execution.

FOUNDER JUDGMENT

YOU CANNOT DELEGATE JUDGMENT

As CEO, you eventually face decisions where advice becomes limited or contradictory.

You still have to choose what makes sense to you.

At high stakes, there may be nobody else who fully holds the context. The quality of your judgment becomes part of the company’s edge.

AVERAGE IS NOT GOOD ENOUGH

Copying the average answer pulls you toward average outcomes.

Successful startups are exceptions.

To create an exceptional outcome, your judgment must sometimes differ from consensus—and be better.

You only learn whether it is better by making real decisions with real stakes.

ACTION PRODUCES INFORMATION

When founders are stuck, more abstract thinking can stop helping.

Action creates new information.

Ship something. Change the process. Run the experiment. Talk to the customer. Remove the blocker.

The available action space is usually larger than it first appears.

WHEN STUCK, INJECT ACTION

Organizations can settle into local minima where everything feels constrained.

Sometimes the next phase only appears after a meaningful change in people, process, product, or structure.

The principle is practical: when the system stops producing learning, change the system.

AI-NATIVE INFRASTRUCTURE

AI IS A MULTIPLIER, NOT JUST A CHATBOT

Science puts company activity into Helix so AI agents can work across real operational context.

The value is not only automation.

It is connecting purchasing, quality, manufacturing, regulatory work, and internal knowledge so AI can assist inside the actual operating system.

WHERE AI ALREADY HELPS

Hodak highlights two major areas:

1. Coding — AI now handles a large share of software creation.

2. Regulation — AI can identify standards, organize evidence, and reduce months of compliance work.

AI is especially useful where humans struggle with huge amounts of structured detail.

CUSTOM INTERNAL SOFTWARE IS BECOMING RATIONAL

Historically, custom internal software was expensive, so startups bought generic systems.

AI coding changes that trade-off.

A company can now build software around its own workflows—like hiring, purchasing, manufacturing, and reviews—creating a tighter operating system than generic tools allow.

CAPITAL & RUNWAY

RAISE ENOUGH TO RUN THE REAL EXPERIMENT

Some ideas are worth funding with $50 million or zero—but not $5 million.

First determine the experiment that must be run to reach the next value inflection point.

Price the real work required. Raising too little can leave you with an expensive but inconclusive outcome.

FIND THE NEXT VALUE INFLECTION

A huge mission still needs a nearer milestone.

Ask:

1. What must become true next?

2. Which experiments prove it?

3. What people, time, tools, and materials are required?

4. How much capital gives the team enough room to complete that cycle?

PUSH TOWARD REVENUE EARLIER

Even long-horizon deep-tech companies benefit from revenue and sustainability.

Fundraising repeatedly resets the clock, but it does not remove dependency.

Revenue gives the company more control, expands the investor base, and shifts valuation toward long-term opportunity instead of survival risk.

TALENT & CAREER

INDUSTRY VS PHD

If the frontier only exists in academia, a PhD can make sense.

But when a field is already moving fast in industry, a high-performing company may offer more resources, faster iteration, and direct exposure to execution.

The answer depends on where the real frontier currently lives.

LOOK FOR EXCEPTIONAL EVIDENCE

Hodak looks for something specific that separates a person from peers.

Competitive achievements are especially legible: winning technical competitions, becoming a chess grandmaster, building and racing systems, or repeatedly doing difficult things with measurable outcomes.

EVALUATE THINKING, NOT ONLY CODING

For software roles, Science increasingly emphasizes how candidates think.

Can they decompose an unfamiliar problem?

Can they identify constraints and interfaces?

Can they reason clearly in ambiguity?

Code generation is becoming cheaper. Clear thinking remains scarce.

WORK NEAR EXCELLENT JUDGMENT

Hodak describes startup judgment as something learned through repeated real decisions with stakes attached.

Working inside a high-performing company can train those filters.

You see difficult situations, watch strong operators decide, then observe later whether the judgment was correct.

SMALL TEAMS, BROAD CONTEXT

INTERDISCIPLINARY MEANS MOVING THE BOTTLENECK

Science prefers smaller teams that understand more of the whole system.

That lets them move constraints across disciplines.

A hard electronics problem might become easier through better biology. A material limitation might be solved through a different architecture.

The system matters more than the org chart.

HANDS-ON DEPTH + BROAD UNDERSTANDING

For complex fields like neural interfaces, start with a real technical capability: software, electronics, mechanical engineering, materials, biology, or another hard skill.

Then widen your understanding of the total system.

Hands-on experience creates the intuition needed to connect disciplines.

THE STARTUP OPERATING SYSTEM

1. Shorten iteration cycles.

2. Build infrastructure before chaos forces it.

3. Move spending control into budgets.

4. Systemize hiring and feedback.

5. Make company context available to AI.

6. Fund real experiments.

7. Act when stuck.

8. Keep sharpening your own judgment.

THE FINAL LESSON

There is no universal startup algorithm.

At the frontier, every company faces a new sequence of facts.

Your advantage comes from building systems that produce faster information, then using differentiated judgment to act on it.

Average decisions create average companies. The frontier demands better.