A protein is an excellent piece of machinery, provided you want the job it was built to do. Evolution had no reason to make a molecule that knows which tumor cells to approach, how long to stay in the body, or whether a patient would rather swallow a pill than book an infusion. AI Proteins begins with that awkward fact. In a Boston laboratory, it designs small proteins from scratch and asks whether a purpose-built molecule can do what a borrowed one cannot.
- AI Proteins designs de novo miniproteins for potential drugs, then makes and tests them in an automated lab.
- It develops its own preclinical programs and works on targets for pharmaceutical partners.
- Its Bristol Myers Squibb agreement carries up to $400 million in possible milestones, not an upfront price tag of that size.
- The company's reported binders span more than 150 targets; turning those hits into approved medicines is the test ahead.
Start with the job description
Most biologic drugs begin with a natural protein and modify it. Antibodies, for example, have changed medicine. They also arrive with a shape and history set by the immune system. Chris Bahl, AI Proteins' founder and CEO, argues for reversing the order: describe the desired medicine, then design a molecule to meet the description. Its small synthetic proteins can be tuned for binding, stability, size and arrangement. Several can be linked or attached to other payloads, making them candidates for tasks that ask one molecule to recognize more than one target.
The distinction sounds almost literary: nature is the author; the biotech wants to be the editor. The important part is less romantic. A drug needs to bind the right target, avoid the wrong ones, survive manufacturing, travel to the right tissue and behave predictably in a body. A protein that looks elegant on a screen can fail at any of those stops. AI Proteins therefore treats a predicted design as a proposal, not a result.
Bahl's route to the company passed through David Baker's Institute for Protein Design in Seattle, then the Institute for Protein Innovation in Boston. The company launched publicly in 2022 with an $18.2 million seed round co-led by Cobro Ventures and Lightchain Capital. That money was aimed at a high-throughput platform: a way to turn a scientist's one-off act of design into a repeatable process. The announced $41.5 million Series A in November 2025 brought the two disclosed rounds to $59.7 million.
The machine has a feedback habit
The platform's sequence is simple to state and difficult to run. Generative AI proposes new proteins. Automation makes many candidates in parallel. Lab assays measure their structure, physical properties and function. Experimental data then goes back into the models; the company calls one part of this optimization system AWESSM. The loop matters because machine predictions are only as useful as the measurements that correct them. It is also the part another research team could copy in principle: define the intended drug profile first, build a fast physical test, and make every test teach the next design round.
By late 2025, the company said it had generated molecules against more than 150 targets and achieved multiple proof of concept results in animals. That is breadth, with a crucial qualifier. A binder is a molecule that attaches to a target. It is not automatically a drug, and an animal result is not a clinical outcome. The company's published pipeline lists internal work on TNFR1 for inflammation and targeted radiotherapy approaches in oncology, along with partnered programs. These are discovery and development efforts, not marketed products.
A pill is harder than a pitch
One of the sharper tests of the idea began in January 2024. AI Proteins and Vivtex agreed to study whether a designed miniprotein aimed at TNFR1, an inflammation-related receptor, could become an oral biologic. The companies said they would share the research costs and the resulting data; they did not disclose a dollar figure. Vivtex brings a gut simulation and formulation platform, while AI Proteins brings the molecule. The aim is practical: a protein cannot help as a pill if digestion destroys it or too little reaches the body.
The collaboration also shows what the company does when it meets a problem outside its own specialty. Its miniprotein can be designed for durability, yet oral delivery is an entire discipline. A partnership allows the team to test that missing piece instead of assuming the design has solved it. There is no announced approved oral drug from the project. For now, the useful fact is the experiment's shape: a specific target, a specific delivery problem and two groups with complementary tools.
“Natural proteins evolved under selective pressures in the environment. They did not evolve to be modern medicine.”Chris Bahl, 2025 Series A announcement
A second collaboration, with the University of Missouri, explores miniproteins carrying radioactive payloads to cancer cells. The researchers are studying whether one designed molecule can recognize several tumor antigens at once. Here the small size and modular nature of the protein may be useful; the open question is whether those properties improve targeting and outcomes when tested rigorously. The university agreement was announced in January 2025 as research, not a clinical success.
What the $400 million really says
The customer who gives the platform its most recognizable commercial validation is Bristol Myers Squibb. In December 2024, the drugmaker signed a research collaboration and option agreement for AI Proteins to discover and optimize miniproteins against two undisclosed targets, with options for two more. AI Proteins receives an upfront payment whose size was not disclosed. The widely repeated “up to $400 million” refers to possible development, regulatory and commercial milestones, plus potential royalties on sales. Those payments depend on future decisions and results.
Still, the deal clarifies the business. AI Proteins is not selling a browser-based design tool or a finished prescription medicine. It is selling access to a discovery capability and seeking to own valuable drug programs of its own. The partnership gives a large pharmaceutical company a way to try the modality on selected targets, while AI Proteins keeps improving the engine. The 2025 Series A announcement introduced a hub-and-spoke structure: the core platform remains central, and specialized programs can sit in separate subsidiaries. That is a way to finance and manage multiple drug bets without reducing the company to a single asset.
The approach puts AI Proteins alongside computational protein designers such as Generate:Biomedicines and Absci, as well as antibody developers and peptide specialists. Its distinctive claim is the combination of de novo, compact modular molecules with automated, high-volume wet-lab testing. No one gets to skip the comparison that matters most: a candidate must be safer, more effective or more practical for a particular clinical job than the treatment a doctor already has.
The laboratory gets the last line
The most revealing thing about the company may be a comment from Frank Teets, its head of computational science. Describing how the team assembled a curated dataset, he said everyone wanted insights but nobody wanted to do data entry. Their answer, he said, made data collection part of the workflow so the whole company could contribute. It is a small operational detail with large consequences. A discovery platform learns from experiments only if those experiments become usable data.
This is where the wager becomes legible outside biotechnology. AI Proteins took a field famous for individual scientific craft and tried to give it a production rhythm. First specify the job. Then make many candidates, measure them in the real world and let the next design remember the result. It works only while the measurements are honest and the next stages - delivery, safety, manufacturing and clinical efficacy - cooperate. Biology is an exacting editor. AI Proteins has built a faster way to submit drafts.
For prospective partners: the company invites discussions through its pipeline team about target-specific discovery. For everyone else, these programs remain research; there is no AI Proteins therapy to request from a physician today.