Max Hodak: What Really Kills Deep Tech Startups?

Max Hodak opens by discussing his background in brain-computer interfaces over nearly 20 years, from early neuroscience research at Duke to his current role at Science, which has developed a retinal prosthesis that restores vision to blind patients through implanted solar-cell chips powered by projected infrared light.

The core thesis of his talk is that infrastructure—not technology—determines success or failure in deep tech companies. He uses the Picasso quote about artists discussing where to buy cheap turpentine to argue that successful founders focus on logistics and operational systems rather than high-level strategy.

On purchasing and spending: Hodak explains the problems that emerge as companies scale. Founders initially approve all purchases but this becomes impossible with multiple employees. Handing out corporate credit cards causes confusion about spending authority. Setting up procurement systems creates delays through B2B vendor processes requiring quotes, purchase orders, and invoices. The solution requires building internal software (Helix at Science) that gives teams budgeting clarity so they understand available resources and make trade-offs appropriately. He notes that Science discovered individual wafer iterations cost $40,000—critical information for runway calculations that many founders don't systematically track.

On hiring and recruitment: Hodak describes a four-step process Science uses: (1) online applications with company-wide voting by employees with similar backgrounds to applicants (distributed to avoid bottlenecks), (2) phone screens assessing judgment, horsepower, and agency, (3) take-home homework that is AI-resistant and naturally scorable, and (4) final interviews. He emphasizes that 17% of top-funnel applications proceed to phone screen, and at least 25% of interview candidates should convert to offers to avoid wasting time on onsites.

On performance reviews: Rather than traditional annual HR-driven 360 reviews, Hodak developed "Eigen Reviews"—a continuous system where employees periodically rate whether they'd rehire their colleagues today. The system uses eigenvector centrality (similar to Google's PageRank) to weight votes based on the rater's own credibility, creating an unbiased continuous signal of performance. This replaces the traumatic once-yearly review process and surfaces issues within weeks rather than months.

On judgment and decision-making: Hodak argues that founders cannot delegate their judgment. Success requires making decisions that make sense to you even when alone in that conviction, as exceptional companies exist at the statistical tails. He emphasizes that at critical inflection points years into a company with hundreds of millions at stake, there is often no one to ask for advice—the CEO must commit to their judgment. However, he notes that action always produces information and the action space is larger than it appears; getting stuck is rare if you keep injecting effort.

On BCIs and consciousness: Hodak positions neural interfaces not as AI-adjacent but as a longevity/healthcare story. He argues that brain-direct engineering avoids unsolved biology problems and produces dramatic effect sizes—deep brain stimulation patients can go from unable to hold water to writing cursive in 10 seconds. He believes BCIs could help prove consciousness is a practical rather than philosophical problem through future experiments.

On infrastructure and speed: The overarching theme is that iteration speed determines success, and infrastructure determines iteration speed. Companies that learn weekly versus monthly will overwhelmingly outcompete. Speed comes from boring systems: purchasing efficiency, recruiting systems that don't bottleneck, and continuous performance signals. Hodak notes most deep tech failures aren't due to technology not working, but inability to organize hundreds of people and manage physical infrastructure as the organization scales.

On building versus buying software: Hodak argues that custom internal software (Helix) is valuable for deep tech companies because commercial ERP systems don't fit their needs. SpaceX's Warp Speed, Facebook's internal tools, and YC's software all exemplify how companies that build bespoke systems tuned to their operations gain efficiency that off-the-shelf software cannot provide. This became more viable with AI-assisted coding.