The preclinical result offers a test of Lila's autonomous-science pitch, though the company has not published the underlying data.
By Ryan Merket · Published
Primary source: X
Why it matters #
Lila's claim moves autonomous science from workflow demos toward a commercial-therapy benchmark, but preclinical data and human results remain the tests that count.
Jing (@JinghanOng) said on August 28th that three Lila Sciences scientists designed a new CAR-T cancer therapy in a matter of months, then produced preclinical results that beat what Jing described as a leading therapy already used in patients.
The claim is an early test of the central bet behind Lila: that AI systems connected to automated laboratories can shorten the cycle between forming a scientific hypothesis, building a candidate and testing it in the physical world. Jing's two-post thread on X directed readers to a Lila recruiting page, tying the scientific result to an effort to add product leadership in software and applied AI.
Lila has not attached the underlying study to the post. The thread does not identify the cancer target, the three scientists, the commercial therapy used as a comparator, the type of preclinical model or the measurements behind the claimed advantage. That leaves the result at the level of a company-reported research milestone rather than evidence that can be independently evaluated.
That distinction carries particular weight in cell therapy. CAR-T treatments genetically engineer a patient's T cells to recognize and kill cancer cells. The National Cancer Institute says the therapies can produce durable responses in some advanced blood cancers, while also carrying serious risks and a lengthy development and manufacturing process. Preclinical performance is an early step before human trials establish safety, dosage and efficacy.
Lila puts its platform against a commercial benchmark
Lila has been preparing the ground for this claim. In a June 22nd technical post, AI Lab Innovation head Ben Kompa said Lila's platform had generated novel untranslated-region, or UTR, sequences with applications in next-generation CAR-T therapies. Kompa described a closed loop in which AI agents propose candidates, run experiments through Lila's automated facilities, analyze the results and choose the next experiments without waiting for a series of manual handoffs.
The August 28th disclosure goes further by placing a Lila-designed candidate against an existing patient therapy, at least according to the company's account. A commercial comparator gives the experiment more relevance than a benchmark confined to computational predictions. It still cannot answer whether the candidate would perform better in people, remain effective over time or meet the safety and manufacturing requirements applied to cell therapies.
The Food and Drug Administration requires CAR-T developers to assess product design, manufacturing controls and clinical risks. Approved autologous CAR-T products carry warnings covering cytokine release syndrome and neurological toxicities, while long-term monitoring remains part of the regulatory framework.
Geoffrey von Maltzahn's broader autonomous-science bet
Lila co-founder and CEO Geoffrey von Maltzahn has spent years forming biotechnology companies inside Flagship Pioneering, including Generate:Biomedicines, Tessera Therapeutics and Indigo Agriculture. Lila, founded inside Flagship in 2023, applies the same platform-company logic across biology, chemistry and materials science rather than committing itself to one therapeutic pipeline.
Lila says its "AI Science Factories" combine models, software, robotics and laboratory instruments so that agents can design and execute experiments, then learn from the physical results. Its therapeutics platform covers nucleic acids, protein therapeutics, delivery systems and in vivo cell reprogramming, with optimization objectives including potency, durability, safety and manufacturability. Lila labels the work as research use only.
The capital behind that thesis is substantial. Lila closed a $350M Series A in October 2025, bringing its announced funding to $550M. The round included Flagship, Braidwell, Collective Global, NVentures, General Catalyst, March Capital, IQT and other institutional backers. Lila said the financing would fund additional automated laboratories and open the platform to commercial partners.
The CAR-T claim supplies the kind of applied result those backers are paying to see: a small scientific group moving from design to a tested candidate on a compressed timeline. Its value will depend on whether Lila publishes enough experimental detail for researchers to judge the comparison and whether the candidate survives the much longer path from preclinical testing to patients.
For now, Lila is using the result to recruit. The linked Product Lead, Software/Applied AI role, based in Cambridge or San Francisco, calls for someone to build an AI-native system of record for scientific discovery and productize scientific agents, model tooling, evaluation systems and virtual wet-lab software. The listed base salary is $180,000 to $288,000.