Chai Discovery Just Hit $3.8B on De Novo Antibodies. The Author List Is 31.

On July 14, 2026, Chai Discovery announced a $400M Series C at a $3.8B valuation, led by Index Ventures. It is Chai's third round in under a year, pushing total funding above $600M, and every biotech and AI lab in the world just got a new hiring target: the roughly thirty people who have actually shipped a working zero-shot de novo antibody generator.

If you are staffing against this specialty, the "ML plus biotech" LinkedIn search you are about to run will lie to you. The real pool is a rounding error, and it lives in four zip codes.

Why the pool is smaller than one paper's author list

The hireable global pool for de novo antibody design is roughly the size of Chai-2's author list, and probably smaller. De novo antibody design is the task of generating a binder to a target protein from scratch, without a starting template, using a learned model. Chai-2's bioRxiv preprint (July 2025) has 31 named authors. Refolk's index of US professional profiles matching "protein design antibody" surfaces only 44 people total.

That means if half of Chai-2's authors stay put (a reasonable base rate for authors on a company's flagship paper), a competitor going after this specialty is fighting over a pool of 20 to 30 credible external hires. Globally. Not per city.

The reason the pool is this small is mechanical, not cultural. Prior de novo antibody methods reported success rates below 0.1%. Chai-2 hit 16% on 52 diverse targets, validating in under two weeks. That is a 100-fold jump. A jump that big means legacy antibody engineering CVs (phage display, hybridoma, humanization) are not substitutes for the people who built Chai-2. They are a different job.

100x

What Refolk's index actually shows

Refolk's US index shows 44 profiles for "protein design antibody" and 85 for the broader "machine learning protein" query, with heavy clustering in Boston, Seattle, and San Diego. Those two numbers are the whole map for de novo antibody design hiring in 2026.

A few things to read out of this table.

• The narrow specialty is roughly half the size of the broader "ML plus protein" pool. So you cannot upskill your way into it with a generic ML-bio hire; the delta is not training, it is domain reps.
• Chai-2's 31 authors represent about 70% of the entire US pool that surfaces on the exact specialty. One company owns most of the shippers.
• The European sub-pool at this specificity is invisible in the index. Cambridge (UK), Zurich, and Munich all return effectively zero on the narrow query. Remote-EU sourcing is not a fallback plan here.

The top current employers on the narrow query (BigHat Biosciences, Absci, Visterra, Invenra, Eli Lilly) are the poaching list. The broader query adds Genentech, GSK, Monod Bio, Abiologics, and Antares Therapeutics.

The OpenAI biology alumni archetype is real, and tiny

The founder profile that produced Chai is a specific and rare crossover: senior ML engineers from OpenAI, Meta FAIR, or Stripe-caliber infra teams, who then pivoted into structural biology. The relevant sub-cluster numbers in the low double digits worldwide.

Chai's four co-founders map cleanly to it:

• Joshua Meier (CEO): ex-OpenAI, ex-Meta FAIR
• Jack Dent: ex-Stripe
• Matthew McPartlon: molecular design
• Jacques Boitreaud: academic structural biology

The interesting recruiter move is to treat "OpenAI biology alumni" and "Meta FAIR ESM diaspora" as first-class sourcing segments, not one-off resumes. ESM (Evolutionary Scale Modeling) was Meta FAIR's protein language model project; its team is small and largely accounted for. Most of them are now at EvolutionaryScale, Chai, or one of six other named startups.

Anthropic is now hiring against the same cluster. Ten months ago, Anthropic put Eric Kauderer-Abrams in charge of a life-sciences team because it decided biology was the single most important place to put its models to work. Isomorphic Labs pulls from the same pond. So does Xaira. So does Generate Biomedicines. Boolean strings do not resolve this cluster because most of these people do not put "de novo antibody design" in their LinkedIn headline. They put "research scientist" or nothing.

Describing the archetype in plain English is the exact gap Refolk closes: ask for "ML engineers who worked on ESM or AlphaFold-adjacent projects and later moved into structural biology startups" and you get a ranked shortlist, not a keyword collision.

Baker lab IP is fenced. Plan around it.

The obvious secondary talent path (University of Washington's Institute for Protein Design) is partially closed because Xaira has exclusively licensed RFantibody's training code. RFantibody is the Baker lab's de novo antibody design system; David Baker won the 2024 Nobel for the underlying work.

Exclusive licensing does not mean UW IPD alumni are unhirable. It means:

1. Current UW postdocs working on RFantibody are effectively pre-committed to Xaira on the IP that matters.
2. The right map is second-degree: postdocs and PhDs who left the Baker lab before the exclusive license, or who worked on adjacent projects (RFdiffusion, ProteinMPNN, hallucination-based design) that are not fenced.
3. The MLSB (Machine Learning for Structural Biology) workshop author lists from 2023 and 2024 are a better sourcing surface than the IPD website, because they catch the people who trained there and then dispersed.

Boolean strings do not resolve this cluster. Most of these people do not put de novo antibody design in their LinkedIn headline.

Every big pharma is now a competing employer

Pfizer, Lilly, and Novartis are no longer just Chai's customers. They are staffing internal AI drug discovery teams against the same 44-person pool. Treat pharma AI groups as top poaching risk, not backfill options.

The signals are on the record:

• In June 2026, Chai unveiled a licensing agreement with Pfizer for early access to Chai-3 plus a custom model trained on Pfizer's proprietary data. Pfizer is building the muscle to consume that model in-house.
• Chai's public deal list includes Eli Lilly and Novartis.
• Nabla Bio (Harvard/MIT ecosystem, $26M Series A in May 2024) simultaneously announced partnerships with AstraZeneca, BMS, and Takeda worth over $550M in combined upfront and milestones. Those three pharma AI groups are hiring on the back of it.
• Refolk's index already shows Genentech, Eli Lilly, and GSK among top current employers on the narrow query.

Recruiter implication: your loss report should track pharma internal moves, not just startup-to-startup jumps. A senior antibody designer who "left for Lilly" a year ago is now working on the exact same problem you are recruiting for, at 2x cash comp, with clinical assets to point at.

The competitive map, ranked by hiring pressure

The companies chasing the 44-person US pool right now, ordered by how aggressively they are staffing against it:

1. Chai Discovery ($3.8B, $600M+ raised, hiring against Chai-3 and Pfizer custom model)
2. Xaira Therapeutics (launched April 2024 with over $1B, second-largest initial biotech funding ever, holds exclusive RFantibody license)
3. Isomorphic Labs (Alphabet spinout, AlphaFold lineage, aggressive on ML-first drug design roles)
4. Generate Biomedicines (Flagship, long-established, competes on scale)
5. EvolutionaryScale (ESM team spin-out from Meta FAIR)
6. Absci (public, generative antibody platform, appears on both Refolk index queries)
7. Nabla Bio ($550M+ in pharma deals, small team, hiring quietly)
8. BigHat Biosciences, AI Proteins, LabGenius, Archon Biosciences, Genesis Molecular AI (mid-tier competitors)
9. Pharma internal AI groups: Pfizer, Lilly, Novartis, Genentech, GSK, AstraZeneca, BMS, Takeda

If you are one of companies 3 through 9 on that list, you are recruiting into a market where the top employer just marked itself up 3x in seven months (from a $1.3B Series B in December 2025 to $3.8B in July 2026). Equity math against Chai is hard. Focused-mission and speed-to-clinic are your only real levers.

Geographic concentration makes "remote US" a losing req

Open a "remote, US" req for a de novo antibody designer and you will lose to a competitor with a physical office in Kendall Square or South Lake Union. The specialty clusters in four hubs, and the people in those hubs will not move.

Refolk's index breakdown of the 44-profile narrow pool:

• Greater Boston: 4 (Kendall Square gravity, MIT/Broad/Harvard adjacency)
• Greater Seattle: 3 (UW IPD gravity, South Lake Union)
• San Diego: 3 (Genentech, Illumina alumni, established biotech corridor)
• SF Bay Area: 2 (Genentech South SF, plus AI crossover)

Those numbers are small because the specialty is small. But the concentration is the story. If you cannot commit to an office in one of those four metros, cut the req to "ML researcher, protein-adjacent" and take the pool from 44 to 85. You will trade specificity for a fightable market.

For that broader search, Refolk handles the plain-English version well: "ML researchers with protein language model experience, currently at biotech, open to remote or Bay Area" resolves in one query what a Boolean-plus-scroll workflow burns a week on.

What to do this quarter

Four moves that account for the pool's actual shape.

1. Build the 44-person map before you post the req. Named list, current employer, publication history, last job change. Refolk's index gets you there in a query; do not start with a JD.
2. Segment by archetype, not title. "Chai-2 author," "ESM diaspora," "second-degree Baker lab," "pharma AI internal (Pfizer/Lilly/Novartis)." Each of those needs its own outreach angle.
3. Show up at MLSB and SynBioBeta. The MLSB workshop is where the shippers publish before they update LinkedIn. SynBioBeta 2026 already featured both Kauderer-Abrams (Anthropic) and Marc Tessier-Lavigne (Xaira CEO). If you are not in those rooms, you are recruiting six months late.
4. Do not bet on Europe as a fallback. UK, Switzerland, and Germany return zero on the narrow query in Refolk's index. Cambridge (UK) and Zurich have great ML-bio talent generally, but not this specialty, not yet.

The market is priced. The pool is not. The recruiters who win this year are the ones who accept that a 44-person map is the actual game board and stop pretending LinkedIn's filter panel can find people who never wrote "de novo antibody" in their headline.

FAQ

How many people can actually do de novo antibody design?

Refolk's US index shows 44 profiles matching "protein design antibody" and 85 on the broader "machine learning protein" query. Chai-2's bioRxiv paper has 31 named authors, meaning one company employs roughly 70% of the surfaceable US narrow-specialty pool. The global hireable pool for outside companies is realistically 20 to 30 people, concentrated in Boston, Seattle, San Diego, and the Bay Area.

Why doesn't LinkedIn keyword search work for this?

Two reasons. First, most shippers do not put "de novo antibody design" in their headline; they write "research scientist" or leave it blank. Second, the archetype crosses two industries (frontier AI labs and structural biology) that LinkedIn treats as different filter trees. Refolk resolves the archetype in plain English instead of Boolean, which is why it surfaces the 44-profile narrow pool as one query.

Who are Chai Discovery's real competitors for talent?

For AI-first startups: Xaira Therapeutics, Isomorphic Labs, Generate Biomedicines, EvolutionaryScale, Absci, Nabla Bio, BigHat Biosciences, and a handful of mid-tier names like AI Proteins, LabGenius, and Archon. For internal pharma AI groups: Pfizer (now running a custom Chai-3 model on its own data), Eli Lilly, Novartis, Genentech, GSK, AstraZeneca, BMS, and Takeda. Anthropic's new life-sciences team, led by Eric Kauderer-Abrams, is also hiring against the same cluster.

Is UW's Institute for Protein Design still a talent pipeline?

Yes, but with a caveat. Xaira holds an exclusive license to RFantibody's training code, which means current Baker lab postdocs working on that stack are effectively pre-committed to Xaira on the IP that matters. The recruitable IPD population is second-degree: alumni who left before the exclusive license, or who worked on adjacent projects like RFdiffusion and ProteinMPNN. MLSB workshop author lists are a better index than the IPD roster.