Ex-DeepMind scientist launches LatentX for protein design

Designing a brand-new protein used to require a team of computational biologists, a cluster of GPUs, and months of setup. Now, you can do it right inside a web browser.

Latent Labs, founded by former DeepMind AlphaFold design lead Simon Kohl, just launched LatentX. It is a web-based foundational AI model aimed at programming biology from scratch.

Generative biology right in your tabs

Most people know AlphaFold for cracking protein folding. But as Kohl pointed out, AlphaFold is primarily a predictive model that visualizes what already exists in nature—it does not generate entirely new molecules.

LatentX takes the generative leap. It designs molecules with precise atomic structures that go beyond natural evolution, aiming to speed up therapeutic discovery.

According to Kohl, the model achieved state-of-the-art results for protein binding in lab benchmarks. Latent Labs claims a high percentage of designs hold up when tested in real-world wet labs.

Best of all, academic researchers, biotech startups, and pharma scientists can access the tool straight from their browser without managing complex local infrastructure.

Selling picks and shovels

Unlike drug discovery outfits like Xaira, Recursion, or DeepMind spinout Isomorphic Labs—which focus on developing their own proprietary drug pipelines—Latent Labs wants to be a platform provider.

Kohl noted that most biotech companies simply cannot afford their own AI research teams and massive compute clusters. Latent Labs plans to license LatentX to external organizations instead.

LatentX is currently free to try, with paid tiers planned for advanced capabilities. The company is backed by $50 million in funding from investors including Radical Ventures, Sofinnova Partners, Google chief scientist Jeff Dean, Anthropic CEO Dario Amodei, and ElevenLabs CEO Mati Staniszewski.

Why it matters

Turning frontier biology into a browser-native app is a brilliant distribution move. Independent labs and cash-strapped biotech startups can now test generative protein designs instantly without hiring an army of machine learning engineers.

Focusing on licensing rather than proprietary drug pipelines also derisks the business model. Drug development is notoriously slow and risky, whereas enterprise tooling generates recurring revenue much faster.

That said, wet-lab validation will be the real test. Computational benchmarks are nice, but pharma clients will only pay top dollar once these generated molecules consistently survive physical testing and clinical phases. With competitors like Chai Discovery and EvolutionaryScale pushing open alternatives, Latent Labs has to prove that its browser convenience comes with unmatched biological precision.