Two years ago, Profluent dropped OpenCRISPR-1, the first AI-designed gene editor, into the public domain and then just walked off stage…talk about a mic drop.

The Nature paper that came with it was a flex.

The open-source release was an even BIGGER flex.

The company spent 2024 and 2025 stacking partnerships with Corteva and the Rett Syndrome Research Trust and raising $106 million from Altimeter and Bezos Expeditions to push their protein language models further.

Now Eli Lilly has shown up with a checkbook.

The Berkeley-based biotech just announced a strategic collaboration with Lilly to develop AI-designed site-specific recombinases.

These are ancient enzymes that can drop large stretches of DNA, kilobase-scale, into precise genomic locations.

What’s the big deal with that, you ask? Well, allow me to explain!

Standard CRISPR tools can correct point mutations and small edits beautifully, but they cannot rewrite long stretches of corrupted gene.

Recombinases can.

The catch is that natural recombinases work on a fixed sequence, so designing custom ones for therapeutic targets has historically been a slog.

Profluent’s thinks that this is a problem AI is uniquely positioned to solve. Their ProGen3 family of foundation models, trained on the company’s Profluent Protein Atlas with billions of sequences, can design proteins from scratch and adapt to lab feedback at a speed wet-lab evolution cannot touch.

Under the deal terms, Lilly gets exclusive rights to advance selected recombinase candidates into clinical development and commercialization.

Deal is worth up to $2.25B – depending on Profluent’s ability to meet development and commercialization milestones.

Profluent retains the platform – duhh – Lilly gets the molecules.

This matters because Big Pharma has been very, very picky about which AI design platforms they license versus which ones they politely admire from a distance.

Lilly already has a serious genetic medicine effort with their Loxo Oncology pedigree and the Verve Therapeutics work in cardiovascular base editing. So, picking Profluent for kilobase-scale editing is a vote of confidence that says, we think this team can deliver something we can put in patients.

The diseases this opens up are the “genetically ugly” ones.

Conditions where dozens or hundreds of different mutations across a single gene cause the same disease, so going mutation-by-mutation with traditional CRISPR is hopeless.

Cystic fibrosis with its 2,000+ CFTR mutations comes to mind.

So does Duchenne with its endless dystrophin deletions.

If Profluent and Lilly land even one program in IND-enabling work, the precedent is significant. AI protein design will have moved from “fascinating research output” to “we sold the IP for clinical development.”