He was born July 22, 1980, in Arlington, Texas — the fourth consecutive generation of his family to take up engineering. The discipline was inherited; the instinct was self-taught. As a kid he ran rooms with string-and-pulley rigs to switch lights and move objects remotely, and built elaborate multi-component machines out of LEGO. The early training was spatial and strategic — systems thinking before he had the vocabulary for it.

The part that matters for everything downstream is the second half: he paired the mechanical aptitude with a genuine investment in art and the creative process. That synthesis produced a conviction he never abandoned — that engineering is not the analysis of existing systems but a creative act , the construction of an entirely new architecture from scratch.

Thomas Jefferson High School for Science and Technology focused the disparate interests. Inside the Harvard-MIT Division of Health Sciences and Technology, under Sangeeta Bhatia, he picked up the habit that would run his whole career: an “ideas book” — a continuously maintained map of architectural solutions to systemic bottlenecks. Not experiments. Architectures.

This is the load-bearing section of the file, so read it as the seed of everything after.

His doctoral work, under Bhatia at HST, started conventionally enough: polymer-coated gold nano-antennas, engineered 100,000× smaller than clinical ablation tools and coated to evade the immune system — the longest-circulating nano-antennas built to that point. Inject them, let them pool in a tumor’s leaky vasculature via the EPR effect, hit the site with near-infrared, and the gold converts light to heat. In preclinical models, a single injection plus a laser pass eradicated 100% of targeted tumors.

That was the warm-up. The conceptually radical invention was what he called Systems Nanotechnology — and the name is the whole point. Traditional targeted nanomedicine relied on isolated particles each independently hunting a tumor by receptor binding: stochastic, inefficient, lonely. Von Maltzahn instead drew from blood coagulation cascades and the stigmergy of social insects — coordination through environmental modification rather than central command — and built a two-population system.

First, benign scout particles find the tumor and, on mild laser heating, physically unwind local collagen, exposing cryptic binding sites. That’s a flare left in the tissue — a signal written into the environment. Then a second wave of assassin particles, carrying the payload, reads the flare and recruits to the site en masse . The result: a 40× increase in targeted delivery over non-communicating systems.

Hold that architecture in your head: autonomous agents, no central controller, coordinating through signals deposited in a shared environment, amplifying each other’s work. He built it at the nanoscale in 2009. He will build it again, three times, at three higher levels of abstraction. He will build it last out of LLMs and lab robots.

The cleanest way to read von Maltzahn’s portfolio is as a single instinct re-instantiated at ascending levels of abstraction. The companies aren’t a scatter — they’re a staircase.

• Physical layer — engineer inorganic matter to interface with biology. Nanopartz · Resonance Therapeutics · Sienna.
• Biological / ecological layer — stop engineering particles, start engineering living ecosystems. Seres · Kaleido · Indigo · Axcella.
• Informational layer — stop engineering the organism, start rewriting its source code. Sana · Tessera · Laronde · Mirai · Quotient.
• Meta layer — stop engineering the molecule, start engineering the process that discovers molecules. Generate:Biomedicines, then Lila Sciences.

Each rung abstracts the one below it. Each rung reuses the same systems-architecture instinct — distributed, cooperative, signal-coordinated. The file from here follows the climb.

At Flagship — which he joined in 2009, and where he’d become the youngest General Partner in the firm’s history in 2020 — von Maltzahn’s first big move up the ladder was to stop treating the body as a container for engineered objects and start treating it as an ecosystem to be modulated.

Seres Therapeutics (2010, founding CTO, with Noubar Afeyan and David Berry) inverted the antibiotic paradigm for C. diff . Instead of carpet-bombing the gut and deepening the dysbiosis, Seres delivered a purified consortium of beneficial spores to out-compete the pathogen and repair the gut’s architecture. SER-109 became the first microbiome therapeutic in history to post positive Phase 3 data — an entire new modality of medicine, ecological rather than chemical.

Kaleido Biosciences (2013, founding CEO) took the same target from the other side — not delivering new organisms but reprogramming the existing microbiome with engineered glycans, precision prebiotics that steer the ecosystem’s metabolic output.

Indigo Agriculture (2013, founding CEO; originally Symbiota ) ran the identical thesis at planetary scale. If a depleted gut microbiome makes a person sick, a depleted soil microbiome makes a crop fragile. Indigo coated seeds with beneficial endophytes so crops could withstand drought and pathogens without environmentally destructive chemicals — then extended into carbon farming, paying growers to sequester atmospheric carbon into soil. The microbiome thesis, lifted from the gut to the biosphere.

The next rung: stop modulating the organism, start rewriting its instructions. As CRISPR matured, von Maltzahn fixed on its two structural weaknesses — it’s bad at inserting large sequences, and its double-strand breaks risk hazardous rearrangements — and built a cluster of companies to route around them.

Sana Biotechnology grew out of Cobalt Biomedicine (2016, Flagship code FL39, founding CEO, with Jacob Rubens) and its Fusosome delivery platform — engineered vesicles that fuse directly with target cells and dump payloads into the cytoplasm, skipping the inefficient endosomal-escape problem of standard lipid nanoparticles. A 2019 merger formed Sana; in early 2021 Sana ran the largest IPO in the history of preclinical biotech — $675M.

Tessera Therapeutics (2018, founding CEO, then Board Chair) answered Rubens and von Maltzahn’s founding question — what if nature evolved a better way to alter genomes than cutting DNA? — with Gene Writing™ , built on the Mobile Genetic Elements first described by Barbara McClintock. MGEs insert sequences into the genome without the double-strand break, large payloads included. Tessera raised a $300M Series C and partnered with the Cystic Fibrosis Foundation toward a genetic cure.

Laronde (Endless RNA / programmable circular RNA, $440M Series B in 2021), Mirai Bio (delivery and manufacturing), and Quotient Therapeutics (2023, somatic genomics, again with Rubens) round out the informational layer.

Then the abstraction jumps off the molecule entirely. By 2018, computational biology still leaned on physics-based folding simulation — Rosetta and its kin — slow, expensive, and error-prone for de novo design. With Flagship principal Molly Gibson (project FL57), von Maltzahn made the bet that defined the era: the protein-folding problem is a pattern-recognition problem, not a biophysics problem. Point the same machine-learning architectures driving language models at the vast corpus of biological sequence, and you can teach a system the grammar of biology directly.

Merged with Avak Kahvejian’s parallel motif-based effort (FL56), the two became Generate:Biomedicines in 2019, von Maltzahn and Kahvejian co-CEOs. The platform generates entirely novel proteins that exist nowhere in nature but fulfill a precise therapeutic spec — at one point designing binders that engage ten structural sites on a target simultaneously. A $370M Series B in 2021; a board carrying Nobel laureate Frances Arnold and Moderna’s Stéphane Bancel.

If the rhyme with the scaling-laws story (see N°006) isn’t obvious yet: the same architectural bet — statistical pattern recognition will beat hand-built simulation — was being placed, in the same window, in two different sciences. Generate is the protein-domain instance of the wager Kaplan plotted in loss curves.

Top rung. Generate designs molecules in silico but still hands the physical experiments to humans. Von Maltzahn came to see human cognitive bandwidth in the iterative execution loop as the final, civilizational bottleneck on the rate of science.

Lila Sciences (founded 2023, unveiled March 2025; co-founder and CEO, with Flagship’s Noubar Afeyan as chairman) is the answer — and the company states the ambition on its masthead: the world’s first scientific-superintelligence platform. The instrument is the “AI Science Factory” — a fully autonomous lab built for models and robots rather than people. AI with genuine scientific-reasoning capability is wired directly to robotic fluidics and assay hardware, and the system runs the entire scientific method closed-loop: hypothesize, synthesize, assay, interpret, refine — with no human in the iteration. Von Maltzahn describes the factories as a new kind of “body” for a scientific “mind” — what he calls “scientific-method machines.” The first stood up in Cambridge; sites in Boston, San Francisco, and London followed. In stealth, the platform generated genetic-medicine constructs that beat the best commercial therapeutics and produced thousands of discoveries across life, chemical, and materials science. At the 2026 Upfront Summit, von Maltzahn described it designing potent antibodies against two distinct targets in a single hour — a year-plus of conventional work.

Now return to § 02. The scout/assassin architecture was: autonomous agents, no central controller, coordinating through signals written into a shared environment, amplifying each other’s output. The AI Science Factory is that exact architecture, abstracted from tissue to the scientific method. The hypotheses are the scouts; the assays are the flares; the next iteration reads the environment and recruits. He’s been building one machine since 2009. Lila is the version that runs on robots and weights instead of gold and gelatin.

The capital has tracked the thesis almost vertically. A $200M seed in March 2025; a $350M Series A — a $115M extension among it — closed that October at a valuation north of $1.3B, with George Church as chief scientist and Andrew Beam as CTO; $550M raised in all by the close of 2025. Then, in early June 2026, Bloomberg reported Series B talks for roughly $2B at a pre-money valuation near $8.5B, anchored by CalPERS and Nvidia’s NVentures. Flagship Labs ventures, in aggregate, exceed $80B in value. The number that matters, though, isn’t the valuation — it’s what the valuation is pricing.

The thesis underneath Lila isn’t biological — it’s a wager about how intelligence accrues. Richard Sutton’s “Bitter Lesson” is the canon text: seventy years of AI keep teaching the same humbling fact — general methods that lean on computation and search beat human-engineered, hand-coded domain knowledge, every time, given enough scale. In silico, that lesson produced the language models. Lila’s bet is that it holds in the physical world too — that one model trained across biology, chemistry, and materials science at once will out-discover any siloed, expert-curated approach. Don’t teach the machine the rules of chemistry; give it a lab and let it find them.

If that sounds familiar, it should. It is the same wager Jared Kaplan plotted in loss curves (see N°006) — statistical scale and search over hand-built structure — ported from text to test tube. Generate was the protein-domain instance of that bet; Lila is the instance applied to the scientific method itself. Two sciences, one architecture, placed in the same window.

And the wager has a moat the software companies don’t. The frontier models are running out of internet — the usable public corpus is finite and nearly spent. Lila’s answer is not to scrape data but to manufacture it: every closed-loop experiment emits proprietary, high-fidelity physical data that no competitor can lift from a journal. The company says it has already generated more than 10 trillion tokens of scientific-reasoning data, and projects the corpus will rival, then exceed, the entire training set of the frontier LLMs by the end of 2026. The moat isn’t the model. It’s the factory that feeds it.

Inside the company, the cross-domain discoveries no human would have proposed get a name borrowed from the game that started all this: “Move 37 moments,” after AlphaGo’s impossible stone — and Lila reports they have been recurring across every domain since late 2025. Von Maltzahn frames the larger inflection the same way. The ChatGPT moment, he says, was a fascinating “talking Wikipedia” — captivating, not yet professional. What science is approaching now is the “Claude Code moment”: the point at which a capability stops assisting the practitioner and begins matching, then exceeding, the best of them — the sensation of a new species walking into a profession that had always been ours.

A platform this strange still has to be something a customer can buy, and Lila sells access to the loop through two doors. Catalyst is Lab-as-a-Service: a partner’s own scientists get API-level access to Lila’s reasoning engine, Lila Iris, and to the AI Science Factories, converting fixed laboratory capital into on-demand throughput — the company reports up to 900× the experimental-validation throughput of manual workflows on some agentic tasks. Creation is end-to-end incubation: a partner brings a problem space or a biological thesis, and Lila runs the discovery campaign to a finished, validated physical asset — molecules, optimized protocols, a defensible IP package — not a paper. As of late 2025 the platform opened to its first (undisclosed) commercial partners.

The proof is in what the loop has already produced. Two-target antibodies designed in an hour against a year-plus of conventional work. In mRNA, a team who were not mRNA specialists used the platform to design sequences more than 3× as efficient as the Moderna and BioNTech COVID formulations by protein expression — in four months — with expression persisting roughly fifteen days against the conventional day and a half, a tenfold gain in durability. In cell therapy, hundreds of thousands of formulations screened toward generating CAR-T inside the body rather than in an external lab, with efficacy shown in non-human-primate models.

The under-told half of the portfolio is inorganic. Lila has reported precious-metal-free catalysts for green-hydrogen electrolysis — routing around the platinum and iridium whose cost has kept green hydrogen uneconomic at scale — and permanent magnets that approach the coercivity of rare-earth magnets without the rare earths. That last result quietly moves the work out of chemistry and into supply-chain sovereignty, which is the door into the next section.

For anyone thinking about how this company tells its own story: the thing being sold is not a model and not a molecule. It is privileged access to the one machine that manufactures proprietary scientific data faster than the world can scrape it. Every Catalyst engagement deepens the corpus; every Creation asset proves the loop. The narrative and the moat are the same object — which is the rarest thing a strategy can have.

Read the rare-earth-free magnet result one more time and the frame tilts from commerce to statecraft. If a private loop can design around the exact materials a geopolitical rival controls, then autonomous science is no longer only an R&D accelerant — it is an instrument of national resilience. Governments have read it that way.

In late 2025 the White House announced the Genesis Mission, a federal initiative its boosters compare in ambition to the Manhattan Project, directing the Department of Energy to wire the compute and experimental capacity of its seventeen national laboratories into a single national AI-for-science platform. The named industry consortium reads like a census of the moment: the foundation-model labs (Anthropic, OpenAI, xAI, alongside Microsoft, Google, and AWS), the silicon (Nvidia, AMD), and the autonomous-science companies — Lila among them.

The bet is not America’s alone. The European Union has stood up a “RAISE” strategy to pool AI compute and scientific data across member states; China has codified an “AI+” mandate that pushes robotics into state laboratories to accelerate materials and defense research. The premise underneath all of it is the one von Maltzahn states plainly: whoever runs the largest, fastest scientific-method machine sets the pace for everyone else’s medicine, energy, and defense. It is also why a public pension fund and a chip company are anchoring the same multibillion-dollar round — scientific superintelligence has become infrastructure, and infrastructure is bought by states as much as by markets.

This is the Context Jamming coda, and von Maltzahn sits on the exact fault line this publication is built on.

His entire ladder bends toward a single endpoint: remove the human from the loop. The scout/assassin system removed the surgeon. Generate removed the structural biologist’s intuition. Lila removes the human from the iteration of the scientific method outright — that is the stated source of its velocity. It is a coherent, defensible, and possibly civilization-altering bet: that discovery accelerates exactly to the degree that human cognitive bandwidth is engineered out of it.

The MoEA Loop — the architecture this site is built with — makes the opposite bet at the editorial layer. Multiple models orchestrate, consult, and dissent; the human is retained deliberately, not as a bottleneck but as an editorial membrane — the load-bearing surface that holds the thesis while the models do the throughput. No single model commits alone, and the human is the one boundary that can’t be automated away without the whole structure losing its shape.

Both are closed-loop multi-agent systems. Both descend, structurally, from the same stigmergic insight. They diverge on one question: is the human a bottleneck to be removed, or the membrane that gives the loop its meaning? Von Maltzahn is running the cleanest large-scale experiment anyone has built on the first answer. This file is filed from the second.

The architectures know what their architects believe. They always do.

Education. MIT (S.B., Chemical Engineering, 2003; Randolph G. Wei Award). UC San Diego (M.S., Bioengineering, 2005). Harvard-MIT Division of Health Sciences and Technology (Ph.D., Biomedical Engineering & Medical Physics, 2010; advisor Sangeeta N. Bhatia).

Affiliations. Flagship Pioneering — General Partner (youngest in firm history at appointment, 2020); founding CEO / CTO / Board Chair across the portfolio below.

Mentor. Sangeeta N. Bhatia, MD, PhD — pioneer of micro/nanotech for diagnostics and oncology; source of the “ideas book” habit.

Collaborators / peers worth naming. Noubar Afeyan and David Berry (Seres, Flagship). Jacob Rubens (Cobalt/Sana, Tessera, Quotient). Molly Gibson (Generate, FL57). Avak Kahvejian (Generate, FL56, co-CEO). George Church (Lila Chief Scientist). Andrew Beam (Lila CTO).

Portfolio. Lila Sciences · Generate:Biomedicines · Tessera · Sana (via Cobalt) · Indigo Agriculture · Seres · Kaleido · Quotient · Mirai Bio · Laronde · Axcella · Sienna · Nanopartz · Resonance.

Lila Sciences (2023– ). Co-founder & CEO. Cambridge HQ; AI Science Factories in Boston, San Francisco, and London. Chairman Noubar Afeyan; chief scientist George Church; CTO Andrew Beam; co-founders include Molly Gibson and Jacob Feala. Seed $200M (Mar 2025) → Series A $350M, valuation >$1.3B (Oct 2025) → Series B in talks — ~$2B at ~$8.5B pre-money, anchored by CalPERS and Nvidia’s NVentures (Jun 2026).

Honors. Lemelson-MIT Student Prize. National Inventors Hall of Fame Collegiate Inventors Graduate Prize. Randolph G. Wei Award. NSF Graduate Research Fellowship. Whitaker Foundation Doctoral Fellowship. Business Insider “30 Biotech Leaders Under 40.”