Near the end of the August 29 episode of Moonshots with Peter Diamandis, the show paused for a video from a biotech company called GenBio AI. The video described a simulated human cell that a researcher can poke, dose and edit, and that remembers every change along the way. When it finished, Diamandis said, "This is the foothills of longevity escape velocity. I've been waiting for this forever." Alex, one of the regular panelists, went further: "Medicine is cooked... This is what the end of medicine looks like. It looks like a virtual cell."
The rest of the segment is a good illustration of how quickly a working piece of software can turn into a plan for curing everything — and of where the software currently stops.
What GenBio actually announced
GenBio AI introduced AIDO Cell on 18 August 2026 as a general-purpose simulator for cell biology. Two things distinguish it from a model that predicts one outcome from one input.
The first is that it is stateful. A world-model engine plus a control layer keeps the simulated cell's condition intact from one intervention to the next, so a researcher can apply a genetic edit, then a drug, then a second drug, and watch the cell's response accumulate rather than starting over each time. The demo video described the model as remembering every change you make, from DNA and RNA up through protein interactions, structures and where those proteins end up in the cell, all the way to the cell's overall shape.
The second is that several different readouts are decoded from that same shared state: gene expression, protein interactions, protein localization and cell-painting morphology, a microscopy technique that captures a cell's visual appearance. Because the state is a piece of software, a simulated cell can be cloned and its treatment sequence branched, so two courses of treatment can be run from an identical starting point and compared.
AIDO Cell 1.0 demonstrates this with two prototype cell lines, K-562 and Hep-G2. A companion tool, AIDO Foundry, adapts models to additional experimental data, and AIDO Lab supplies the research interface.
GenBio's own example is a caution as much as a demonstration. It explores molecules that bind ABL1, a protein targeted by the leukemia drug imatinib, and compares their predicted effects with imatinib's. Two molecules can produce a matching transcriptional response — the cell's genes react the same way — while still binding differently or hitting unintended targets. The company reports case studies drawn from published literature and an initial Virtual Cell Benchmark; laboratory validation of the simulator's novel predictions is underway. Timing and dynamics, signaling and metabolites are listed as areas still in development. The immediate use GenBio describes is narrowing hypotheses and prioritizing which experiments to run before committing laboratory resources.
From 10,000 compounds to 10
Diamandis framed the appeal in arithmetic. Cell simulation, he said, could reduce wet-lab experiments a thousandfold: instead of testing 10,000 compounds at the bench, simulate them and test only the top ten the simulator ranks highest. That figure is his extrapolation from the idea, not a result GenBio reports.
He arrived at the segment from a discussion of a personalized mRNA cancer therapy, and his lead-in was about regulation rather than software — the hope that if you can simulate a treatment and show it works in silico, you could "do the studies in a GPU cluster and say, yep, it's safe, let it go." Nothing in the announcement supports that step. Prioritizing which experiments to run in a cell line and establishing that a treatment is safe in people remain separate problems.
Alex's tree search
Alex's version of the endgame borrows from game-playing AI. Train a foundation model to represent all cell states and all interventions against cells, he said, then run "an AlphaGo-type tree search against possible interventions" — exploring branching sequences of treatments the way a Go program explores branching sequences of moves — to find a path that steers a virtual cell from a diseased state to a healthy one. Generalize that from cells to tissues to organisms, and "boom, you've solved all human disease." He compared the move to AGI: you did not have to solve human intelligence, it turned out you could get there by compressing general human knowledge. "It's not that hard in principle."
He declined two of the more romantic readings. Longevity escape velocity — the point at which medicine extends life faster than time takes it away — he expects to arrive earlier and more mundanely, probably through a fourth or fifth generation of GLP-1 drugs; the virtual cell is a superset of that, not the route to it. And when Diamandis described feeding in your own DNA sequence and blood chemistry to learn whether a drug works for you, Alex said he did not want to over-romanticize the personalization. An ideal virtual cell is personalized the way a personalized prompt to ChatGPT is: the output is a function of your inputs, but the thing underneath is a generalist model.
Reading, writing, comprehension
Salim placed the announcement in a longer trend of turning biology into information, which in his account is what puts a field on an exponential curve. Each of us has roughly 50 trillion cells, he said; model that, and a human being becomes a software engineering problem, with the familiar software operations of reading, writing and understanding. Borrowing the vocabulary of learning a language, he said biology has done a good job of reading, has started writing with CRISPR gene editing and mRNA vaccines, and that digital twinning of cells is what finally offers depth on the comprehension side.
Whose data, and who pays
Emad Mostaque wanted the underlying data opened up. This, he said, is what he had hoped a "Genesis project" would be: a Manhattan Project to cure disease, built on in-silico models of cells and whole bodies plus the organized collective knowledge of cancer, autism and the rest. He was blunt that it will not happen that way — "It's not going to be a government program. The labs are going to do that for us" — but argued governments should pool what they hold and make it open anyway, citing UK access as a precedent. Diamandis extended the idea: governments could externalize their data to the private laboratories building foundation models, as a public good they can all train on, a common crawl for biology. Mostaque's case for it was comparative: understanding the body like never before is "much better than building an atom bomb even, because it will have the biggest impact on humanity ever."
The panel expects the frontier AI labs to fund the modeling itself, a point they had argued earlier in the same episode. Diamandis relayed a conversation with Eric, who heads life sciences for Anthropic and told him that his brief from Dario Amodei amounted to unlimited budget to accelerate basic science and cure disease within five years, and to extend human health span within a decade. Amodei's Machines of Loving Grace sets out that ambition as a conditional picture of what could happen if development goes well, with its detailed scenarios presented as educated guesses, and it distinguishes preventing disease from reversing damage that has already occurred. Alex's reading of the strategy was sharper: curing disease is now the best available marketing for not being slowed down in recursive self-improvement, because "you can't slow down the company curing cancer." Mostaque agreed it was good marketing and added a market argument — the biggest market in the world is living another year — while Diamandis argued the work will come from the frontier labs simply because nobody outside them has the compute.
The part that applies outside biology
Dave offered the generalization for listeners who will never build a cell simulator. AI capacity is expanding fast, he said, and it is data-starved. A full cell simulator is a way to design thousands or hundreds of thousands of experiments and get reasonably good test results back through simulation instead of running millions of bench assays — data manufactured where none existed. He pointed to the valuations of companies wrestling new types of data into a form AI can use, citing one of them at tens of billions of dollars with his own hedge about which number was current. Every field of endeavor, he said, is going to be data-starved, and most listeners know a field that needs to supply data back.
The segment closed with a note that the company behind AIDO Cell was co-founded by David Baker, who shared the 2024 Nobel Prize in Chemistry for work on proteins: with structural biology arguably settled, whole-cell simulation is the next grand challenge, and solving it, in the panel's phrase, gets you halfway to solving all disease. Halfway, on GenBio's own account, still runs through the laboratory.