How Infrastructure and Hiring Systems Power Deep‑Tech Startups

Max Hodak, CEO of Science, discussed the critical role of infrastructure in startups, particularly in deep tech companies. He emphasized that while art critics discuss form and meaning, artists focus on practicalities like where to buy cheap turpentine. Similarly, professionals in startups prioritize logistics over grand strategy.

Science: A Retinal Prosthesis Company

Hodak's company, Science, develops a retinal prosthesis designed to restore vision to patients who have lost their rods and cones. This device, featured on the cover of Time magazine, is a chip implanted under the retina. Patients wear glasses with a camera that captures the world and a laser projector that projects an infrared image onto the implant. Where light is absorbed, an electric field stimulates the retina, bypassing damaged photoreceptors and sending visual signals to the brain. The product has completed major clinical trials, was covered by the BBC, and has enabled a patient to read a 300-page novel.

The Importance of Infrastructure

Hodak highlighted that while software companies might operate with minimal physical infrastructure, deep tech companies require extensive purchasing of thousands of items, from microscopes and 3D printers to chemicals and gases. This constant need for supplies presents significant logistical challenges.

Procurement Challenges

Initially, founders might use credit cards for purchases, but scaling this to a team of 17 employees becomes problematic. Approving every purchase individually leads to bottlenecks and inefficient spending. For example, a $3,000 power supply might seem expensive, but delaying its purchase to save money can cost far more in lost productivity from highly paid employees.

Setting up a procurement system introduces its own complexities. Business-to-business (B2B) purchases often involve lengthy processes: - Setting up vendor accounts. - Dealing with insurance and certification paperwork. - Generating quotes, purchase orders, and invoices.

These steps can delay delivery by weeks, frustrating employees and hindering progress. Hodak stressed that efficient procurement is not just about saving money but about enabling speed, which is crucial for a startup's success.

Cost Attribution and Budgeting

Another challenge is attributing costs accurately. When buying materials in bulk for various experiments (e.g., gases, resins), it's difficult to know the exact cost of each experiment or product. This lack of attribution can lead to the perception that experiments are "free," making it hard to manage burn rate and understand profitability.

To address this, Science developed an internal software system called Helix. This system tracks every step of the company's operations, from purchasing to manufacturing, allowing for precise cost attribution. For instance, they discovered that each iteration of a wafer cost $40,000. Understanding these costs is vital for budgeting, managing runway, and making informed decisions about pricing.

Hiring: A Critical Infrastructure Component

Hiring is another area where robust infrastructure is essential. Hodak believes that the best companies often emerge from specific "scenes" or networks. Initial hiring should leverage these networks, as these individuals already understand the company's language and mission. However, to scale, companies must hire from the general public.

Science's Four-Step Hiring Process

Science has developed a structured, software-driven hiring process to manage the overwhelming volume of applications:

1. Company-wide Voting: Applicants apply online, and their resumes are presented to a randomly selected group of 7-8 current employees whose backgrounds are similar to the applicant's. Employees vote on whether they would hire the candidate ("known good," "strong yes," "yes," "no," "strong no"). This distributed voting mechanism prevents bottlenecks and averages judgment across the company.
2. Phone Screen: Approximately 17% of applicants proceed to a phone screen, conducted by a company-wide pool of employees. This screen assesses three key qualities:
   • Judgment: Ability to make good decisions in complex, vaguely defined situations.
   • Horsepower: Basic technical competence and ability to learn.
   • Agency: Effectiveness in achieving desired outcomes and pursuing ambitions.
3. Homework: About half of those who pass the phone screen receive homework assignments. Science prefers "AI-resistant" homeworks that:
   • Don't saturate (have a high ceiling).
   • Are naturally scorable with 2-3 numbers for easy plotting and comparison.
   • Examples include GPU kernel optimization tasks where the goal is to achieve minimum cycle counts.
4. Full Interview: Candidates who excel in homework proceed to a full interview. The goal is to achieve at least a 25% conversion rate from interview to offer, to avoid wasting time on candidates who won't be hired.

Hodak emphasized that this four-step process is the minimum required to make a comprehensive hiring decision efficiently.

Performance Reviews: Eigen Reviews

Traditional 360-degree reviews, conducted annually, are often disruptive and tend to confirm existing knowledge rather than surface new issues. Hodak developed a more continuous and unbiased system called "Eigen Reviews."

Every 4-6 weeks, employees are prompted through the Helix software with a single crucial question: "Knowing how this person turned out, would you vote again today for their hire?" This mirrors the initial hiring vote. The feedback is then used to construct a graph of the company, where an employee's vote is weighted more heavily if they themselves are highly rated by others (similar to Google's PageRank algorithm, an eigenvector centrality concept).

To prevent collusion or malicious downvoting, Science employs techniques like "dropout," where random edges (votes) are removed during iterative calculations. If additional peaks appear in the score distribution, it signals potential "voting cliques" that warrant further investigation. This system provides continuous, largely unbiased feedback, offering better insight into performance and reducing the need for traumatic annual reviews.

Speed and Iteration: The Core of Success

Hodak's central theme is that rate of iteration separates success from failure. A company that learns one thing every week will far outpace a competitor that learns one thing every month. This compounding effect is so dramatic that even if an alternative approach has redeeming characteristics, the one with a shorter iteration cycle should be prioritized.

Speed is determined by infrastructure—how well purchasing, recruiting, and spending processes work. Hodak noted that many deep tech companies, founded by brilliant scientists and engineers, fail not because their technology doesn't work, but because they lack the systems to manage large organizations and physical infrastructure.

The Founder's Judgment

A challenging aspect of being a startup founder is the inability to delegate judgment. Founders must make decisions that make sense to them, even when alone in that realization. Success in startups requires judgment that is "differentiatedly good," not merely average. While advice is abundant early on, critical, high-stakes decisions years into a company's life often fall solely on the CEO, demanding a strong sense of one's own judgment limits.

Action Produces Information

Hodak believes it's difficult to truly get stuck. The "action space" is always larger than it appears. Taking action, even when uncertain, generates information. This is akin to the concept of "action" in physics, where energy exerted creates information. When a company is stuck, injecting action and producing "entropy" can unblock it, sometimes even requiring difficult decisions like removing a strong but ill-fitting employee.

The company's "operating system"—its processes for purchasing, accounting, recruiting, performance reviews, budgeting, safety, and quality—profoundly impacts its potential. Getting these foundations right early makes everything else easier; neglecting them can lead to uncontrolled spending and unwieldy operations.

Q&A Highlights

Industry vs. Academia

For fields primarily existing in basic research, a PhD might be reasonable. However, when a field matures (e.g., computer science, rocket science), industry often marshals greater resources and moves faster. Hodak suggests that if the choice is between working at a high-performing company and getting a PhD, the company might be a better choice for learning organizational skills.

Evidence of Exceptional Ability

Hodak looks for concrete evidence that separates an individual from their peers, ideally winning "legible competitive games" like chess grandmaster titles or Formula SAE competitions. These provide competitive feedback necessary to gauge exceptional ability.

Hiring Engineers

Science doesn't use LeetCode. Instead, they evaluate "thinking" through design questions, assessing an applicant's ability to break down problems and think clearly, rather than just write code.

Lessons from Neuralink

Working with someone with empirically excellent judgment, like Elon Musk at Neuralink, was invaluable for Hodak. This experience provided "reinforcement learning" for making high-stakes decisions, which he believes is an essential part of an entrepreneur's education. He recommends working for a company with an excellent culture before starting one's own.

BCIs and Consciousness

Hodak believes BCIs could absolutely help understand consciousness. He views it as a practical, not philosophical, problem. If a magical BCI could instantly read and drive every neuron, consciousness would be figured out quickly. While current human trials aren't focused on this, it's a long-term goal.

Contributing to BCIs

Neural interfaces are highly interdisciplinary. Hodak advises having a broad perspective of the problem and deep, clear understanding of the system. Hands-on experience and a "hard skill" (software, electronics, materials, etc.) are crucial entry points, followed by continuous learning across disciplines.

AI in Scientific Research

AI acts as a multiplier, not a replacement, for humans. Science uses AI for coding, and significantly for regulatory compliance (quality systems). AI can quickly identify applicable standards and generate evidence tables, streamlining a historically slow and bureaucratic process.

Building Internal Infrastructure

Hodak argues that commercial ERP systems are often disliked, and companies that build custom software tailored to their unique needs (like YC, Facebook, SpaceX/Tesla with Warp Speed) gain significant efficiency. The ability to "vibe code" now makes building internal tools more feasible, allowing for bespoke solutions that commercial software cannot offer.

Impact of BCIs

Hodak sees BCIs primarily as a "longevity story" and a form of healthcare. Neural engineering, including retinal and cochlear implants, offers "effect sizes" rarely seen in medicine, directly addressing the brain as a computer. This approach yields reliable, large effects, improving and extending life.

Preventing Collusion in Eigen Reviews

The Eigen Review system uses techniques like Markov chain Monte Carlo dropout to detect voting cliques. It's also just one of several signals used for performance evaluation.

Funding Deep Tech Startups

Founders should determine the cost of running the next critical experiment (value inflection point), then raise twice that amount. Hodak emphasizes that some ideas require significant funding or none at all, but not an insufficient amount that leads to ambiguous outcomes. He also stresses the importance of pursuing profitability sooner, as it allows the company to be valued on its long-term roadmap rather than its probability of failure.

Best Advice

Hodak believes there are no generic algorithms or shortcuts for startup success. Each situation is unique, and founders must rely on their own refined judgment. He highlights the historical anomaly of being able to raise millions for an idea, emphasizing that this opportunity to push the frontier requires independent thinking.

Hardest Engineering Challenges in BCIs

Key challenges include power and thermal constraints on implants, requiring low-power electronics. Packaging (or "tropicalization") is also critical: developing materials that protect the implant from the body's corrosive environment without harming the body, and can survive for extended periods. This material science problem is highly open-ended.

Interacting with the Medical Field

Science acquired a European company with a retinal prosthesis technology. They actively engage with surgeons and doctors, often through cold emails, to develop new surgical procedures and gather expertise. Hodak noted the cultural blend within Science: 30% are "overtly transhumanist" with a long-term vision, while 70% are "serious clinicians and scientists" focused on building valuable medical devices for unmet needs. This integration of cultures allows them to pursue cutting-edge research while navigating conservative medical regulations.

Breaking into Biotech

Biotech remains capital-intensive and a harder path than other fields, requiring a decade-long commitment. However, the impact on patients and families when successful is unparalleled. Hodak believes that for many effective founders, it's not about money but about changing the world, and a for-profit company is a powerful vehicle for marshalling resources to achieve that.

Popular Beliefs in Tech

Hodak considers the idea that one shouldn't "vibe code a purchasing system" as a contrarian belief he challenges. He argues that building custom internal software, while historically expensive, is now more feasible with AI and can be incredibly powerful when tailored to a company's specific needs.