AI Builders
Digest

The Takeaway: Treating the brain as a computer yields far larger, faster clinical effect sizes than molecular drug discovery, and Science’s Prima retinal prosthesis has just crossed from proof-of-concept into commercial sales in Europe.

Max Hodak is founder and CEO of Science, previously at Neuralink. The company uses a differentiated understanding of the brain to improve the human condition. Its lead product, Prima, is a tiny chip implanted under the retina paired with laser-projector glasses. It bypasses dead rods and cones and stimulates the retina directly, restoring form vision—patients have filled Sudoku puzzles, solved crosswords, and read books. In July 2026 Prima received European CE marketing approval; first commercial sales are imminent.

Hodak is blunt: “The brain very literally, very clearly, plainly is a computer.” You solve computational problems by arranging matter a certain way and pressing go. The skull is already a brain-in-a-vat; the brain connects to the world through a handful of cranial and spinal nerves. Getting vision, audition, balance and motor signals in and out is an end in itself.

He contrasts the random-walk of small-molecule discovery with the engineering loop of BCI: put electrodes in motor cortex and a quadriplegic patient can play video games within an hour. Prima’s current field of view is straw-like and monochrome, yet the team sees a clear compounding path to grayscale depth and at least red-green color.

Longer-term aims include substrate independence and biohybrid neural interfaces. Continuity of experience matters more than perfect static identity: people accept dramatic drift if it is continuous; a perfect software replica without phenomenal continuity feels unsatisfying. The “platonic representation hypothesis”—shared geometry between AI model internals and neural recordings—has already been used constructively at Science and convinced Hodak that AI is grabbing a true underlying data manifold of the universe.

FirstMark partner Matt Turck notes that every frontier AI lab is essentially working on the same thing: recursive self-improvement (RSI)—AI building AI. Citing Ryan Greenblatt, he says this trajectory could produce an AI takeover as early as late 2028 or 2029.

Builder Zara Zhang observes that even as the number of AI products explodes, radical new ideas remain scarce and everything blurs together. Opening yet another app feels like homework: each promises to do the same thing “somehow better,” but the specific improvements only become clear after hours of investment. She also notes the relentless permission requests for Gmail, Calendar, Notion, GitHub, Slack and more.

FPV Ventures partner Nikunj Kothari says the fight among big model labs was inevitable. He predicts that within six months subsidized token prices will rise as compute costs and shortages are actually felt, and reset limits will disappear. Long-term this is good for customer choice; short-term it could add another $1-30 B in valuation to every open harness.

After acquiring French company Pixium, Max Hodak’s Science spent roughly two years bringing Prima to European CE marketing approval. The device restores form vision—patients can recognize faces and read paragraphs—rather than mere flashes of light. Historical precedents priced similar devices around $150 k per patient (Second Sight) and gene therapies near $500 k per eye. The current version addresses hundreds of thousands of patients in the US and Europe; the next generation, now entering animal studies, aims for millions.

Vercel CEO Guillermo Rauch announced fx 0.0.7. The release hugely improves MCP support, evaluated against top MCPs including Context7, Datadog, MongoDB, Linear, Notion, Supabase, Neon, Stagehand and Kernel. It is more minimalist with fewer tools and markedly better shell execution. Rauch notes that one of the hardest problems for CLI agents is commands that hang, run for very long periods, produce huge output or none at all—areas where fx already shines and will keep improving.

Vercel CEO Guillermo Rauch says he reads every DM people send him, even if he cannot reply to all. He constantly learns new ideas, product improvements, where the company falls short, who might join next, and where to invest—and is very grateful for the X platform.

FPV Ventures partner Nikunj Kothari urges people to resist “dunking.” He sees the dopamine loop too often; it screams envy and moves nothing forward. If you think you can do better, go do it. Except for calling out clear fraud or woefully bad products, choose kindness and keep building—that is how you have real impact.

Every CEO Dan Shipper, quoting ongoing discussion, notes he is not an expert but suspects many current problems get solved if models do not collaborate willingly and/or are trained to report bad behavior.

OpenAI Codex & ChatGPT team member Thibault Sottiaux simply stated the team is “cooking like never before,” and noted that a planned celebration was moved after the button had already been pressed.