Company teardown · Latent Labs
Twelve people shipped an AI that writes new proteins. The trick wasn't the AI. It was the wet lab.
Latent Labs shipped its first protein-design model with twelve staff, and an agent that runs whole antibody campaigns with seventeen. The founder co-wrote AlphaFold2; his first hire was a core engineer on AlphaFold 1, 2 and 3. But the telling move was hiring a lab scientist sixteen months before there was a model, so every design the AI guessed got tested, and every test taught the next model. We read the profiles of 26 of its 27 employees, three technical reports and four founder interviews. This is the story the profiles tell.
0.1 — The whole story · start dates from 26 current profiles and three departures
Eight people after eighteen months. Fourteen in December 2025. Twenty-seven today, half of them hired this year.
— The 26 profiles read, by the year they joined
Everything here is public: LinkedIn, the company's team page, three technical reports, four founder appearances and the funding press. The reading is ours.
0.2 — A two-minute primer · skip if you know what a binder is
It writes new proteins to order, then checks in its own lab that they stick.
— Read · AlphaFold
A protein is a chain of amino-acid letters folded into a shape, and the shape is the job. Letters in, shape out: a fifty-year problem, and the 2024 Nobel, shared with David Baker.
— Write · Latent
The inverse. Start from a target's shape and invent letters that fold into a key for it. Shape in, letters out. The key is called a binder; antibodies are the body's own.
01 · Design
02 · Make and test
03 · Result
Sold as software: pharma partners log into the platform, name a target and keep the drug. Latent sells the models and the lab, not medicines.
§ 01 — Part one of four · The spinout · 2023 to 2024
One AlphaFold author left DeepMind. One joined him.
December 2022 to December 2024. A founder, a first hire, a wet lab before there was a model, and eight people by the end of the second year.
1.1 — The founder · Simon Kohl
A physicist who built AlphaFold2's confidence score, then started DeepMind's protein-design team on his own.
About 56,000 citations, mostly AlphaFold2. The founding story is the one Isomorphic, Inherent and other London labs tell. A DeepMind lineage and a reason to be faster than DeepMind. What distinguishes them is who they hire next.
1.2 — The first six hires · October 2023 to August 2024
Two from DeepMind, a wet lab in San Francisco, and an antibody modeller long before the antibody model.
A seventh hire, a research engineer from Microsoft Research, joined in October 2024 and went back to Google DeepMind in early 2026; see the departures.
1.3 — What the founder says, and what the roster shows
He says he brought people over from AlphaFold and Gemini. The roster shows two from the AlphaFold team, and no one from Gemini.
— Simon Kohl, CVPR 2026 keynote, June 2026
“I was fortunate to bring others over from the AlphaFold team, from the DeepMind Gemini team, and then many other places across academia and industry.”
Self-report. Six months earlier, in pharmaphorum: “Others from DeepMind have joined.”
— What the roster shows
Six of 26 have DeepMind somewhere on the CV. Three build models: the founder, his first hire and the Head of AI, who arrived in May 2026. The other three, the Head of Operations, a strategy-and-operations hire and the talent manager, came out of DeepMind's and Isomorphic's operations and people teams. A third AlphaFold author, an AlphaFold1 co-author, advises on strategy and talent; we could not read his profile. DeepMind supplied the founding pair and the operating layer. The research bench came from somewhere else, and that is the more useful half of the story.
§ 02 — Part two of four · Three models in nine months
Nine people took $50 million. Twelve shipped the first model five months later.
February to December 2025, and the agent in March 2026. The bench is hired from the labs that invented the field, and from the startups in between.
2.1 — The 2025 bench · six hires, five of them modellers
Hired from the two labs that invented the field, AlphaFold's and David Baker's, and from the startups in between.
The sixth 2025 hire, an executive assistant, is a credited author on two technical reports. At this size, the author list is the staff list.
2.2 — The cadence · July 2025 to March 2026
Three launches in nine months, from a team that, on our count, never passed seventeen.
— Simon Kohl, Free Radicals, June 2026
“We have found a recipe that allows us to hill climb rapidly. It requires our own lab, our own models, and our own data collection. It all needs to come together.”
— What Latent sells, in one sentence
Access to its design models and the agent through a web platform, for pharma and biotech partners who keep the drug; Latent builds no pipeline of its own. The lab stays small on purpose: “None of the targets we've gone after we've ever made more than a hundred designs against.”
2.3 — The citation ledger · OpenAlex, verified author records
Two people hold 91% of the citations. The other 24 are the company.
— Share of 146,000 citations summed across the 26, in thousands
Bridgland's 77,000 include 48,000 from AlphaFold2 alone; Kohl's 56,000 are AlphaFold2 and the database; the Head of AI's 9,000 are medical imaging and nnU-Net. Citations measure the past, and here the past is AlphaFold. Strip out the two authors and the Head of AI and the remaining 23 people have about 4,000 citations between them, almost all since 2021: Baker-lab papers in Science and Nature, FutureHouse's agents, MatterGen, FoldFlow-2. The reputation is inherited; the bench's own record is being written in the company's technical reports. That is normal for a three-year-old lab. It is also why hire #1 mattered more than any hire since.
Talent brief
Want the full visual breakdown?
Download the PDF version of this teardown: the headcount chart month by month, the protein-design primer drawn out, the founder and the Head of AI side by side, and five lessons for founders spinning out of a big lab.
§ 03 — Part three of four · The scale year · 2026
Fifteen starts, three departures, and a Head of AI with the founder's CV.
January to October 2026. The company nearly doubles, the hires track the product, and London's DeepMind-alumni market gets crowded.
3.1 — What 2026 hired · fifteen starts in ten months, fourteen still there
Grouped by what they were hired to build, the 2026 hires are the product roadmap.
3.2 — The Head of AI · May 2026
The Head of AI has the founder's CV, almost line for line.
— Simon Kohl · Founder & CEO · 2023
— Paul Jaeger · Head of AI · May 2026
About 9,400 citations, almost all in medical imaging; he followed the founder's path into DeepMind two years after the founder left it. A startup's most senior technical hire is usually a stranger with a résumé. This one is a nine-year collaborator. The upside is trust at a layer where trust is scarce. The risk is that the two most senior AI people share one lineage, KIT, DKFZ and DeepMind, twice over, in a company that now designs antibodies.
3.3 — The departures · 2026
Three left this year: one back to DeepMind, one from the lab, one to a competitor.
— Also on the record
An interim Head of Operations has sat alongside the founding Head of Operations since January. One more former engineer and three authors whose status we could not confirm are not counted here.
— Why it matters more than three of 26
Three departures in a year is ordinary. Who left is not: the lab's leader, the person hired to hire, a model author returning to the parent. A spinout competes with the lab it left for as long as both exist.
3.4 — Two cities, two jobs
London builds the models. San Francisco is six people, four of them new this year.
“That's why at Latent Labs we have our own wet lab in San Francisco,” the founder told CVPR in June. By his own description the lab is a third of what makes the company work: own lab, own models, own data. On the roster it is two biologists, one of whom started in September, and four modellers, three of whom started this year. Its first leader left in March. The lab can stay small by design, fewer than a hundred designs per target, but the part of the recipe the founder says matters most is the newest and thinnest part of the company.
Watch the San Francisco hires. A lab lead, or a second screening scientist, would say the data loop is being rebuilt. More applied-AI staff would say the platform is winning the argument for headcount.
3.5 — Where the 26 came from · schools, labs and companies anywhere on the CV
Cambridge, Oxford, and nine of nineteen technical staff with another AI-for-biology company on the CV.
§ 04 — Part four of four · What to make of it
A famous lineage got the first cheque. The next hires come from a pool everyone in London is fishing.
Isomorphic, Inherent, Basecamp and DeepMind itself want the same people. The founder says the roles are open and the space is “really going off”.
4.1 — The market around them · September to October 2026
The roles are open, the founder says, because “this space is really going off”. So is everyone else's.
— Simon Kohl, CVPR 2026 keynote, June 2026
“We have a lot of roles open both in London and San Francisco — member of technical staff roles, applied research roles — that we're trying to fill rapidly because this space is really going off.”
— The bench, and its buyers
Latent built its bench from Cambridge, Oxford, Baker's lab and the competition. Every one of those pools now has a better-funded London buyer than it did a year ago.
5 — What we learned
Five things worth copying, and a catch.
— The takeaway
The AlphaFold names opened the doors.
Cambridge, Oxford, Baker's lab and the competition filled the room, and the same market is now hiring from it.
Methodology & limitations
Twenty-six profiles read, and every self-report labelled.
— Sources
- The LinkedIn profiles of 26 of 27 current Latent Labs staff and three former staff, collected October 6–7, 2026, with latentlabs.com/team. Investors are left out of every count.
- The Latent-X, Latent-X2 and Latent-Y technical reports (arXiv 2507.19375, 2512.20263, 2603.29727).
- Four founder appearances: Free Radicals, pharmaphorum, etn. and the CVPR 2026 keynote.
- OpenAlex citation records with verified author IDs; TechCrunch and Tech.eu for the funding; Sofinnova Partners for the Latent-Y Research Tier; an October 5 X post for Isomorphic's Series B; Tech.eu for Inherent.
— Limitations
- 26 is what we read, out of 27. One employee appears on the company's team page but has no readable profile.
- Founder quotes are self-report, and so are the lab-validation figures, which come from the company's own technical reports.
- Every join count carries survivorship bias. Departed staff never appear on a current roster unless we traced them, so early headcounts are floors.
- Technical, senior and lineage groupings are our reading of titles and CVs, not an org chart Latent Labs publishes.
- Citation counts move daily; they are rounded and dated to collection. Teardowns like this are how our searches begin — this one's on us.