Labor Replacement Is AI’s Small Prize

Author: Arnold Engel (guest post in The Leverage, hosted by Evan Armstrong) URL: https://theleverage.substack.com/p/labor-replacement-is-ais-small-prize One-line: “Much of tech looks at AI as labor replacement, which misses the exploration work that is being scaled up right now that will come to dwarf it.”

Key claims

• The standard AI market-size thesis — AI replaces software engineering, legal work, accounting — is bounded by the actual quantity of that work humans want done. There is no demand for 1,000x more legal work. Scientific discovery has no such ceiling: it is fundamentally unconstrained by existing labor TAM.
• The science flywheel has three stages: Aim (choose research direction), Meet the real world (run physical experiments), Learn (update the model from verified data). Traditional LLMs can do the first and third steps but cannot close the loop without wet-lab contact; that verification gap is now being closed.
• Physical Superintelligence ($58M raised) has human scientists set research direction while AI agents execute experiments. When Astra (OpenAI) launched it set a baseline: operating lab instruments at roughly 100x the pace of PhD graduate students.
• Lila ($550M raised, spun out of Flagship Pioneering) is the breadth bet: “breadth gives depth faster.” Training across chemistry, biology, and materials science produces cross-domain benefits — polymers and proteins are both just large molecules, catalysis and enzymology run on the same underlying physics. Lila has trained on only 2% of its collected data (compute-limited) and claims a proprietary dataset of ~10 trillion tokens of scientific reasoning from verified experimental data, plus 10,000+ evaluation questions across science subdomains worth an estimated $100M on their own.
• Periodic Labs ($300M raised) is the depth bet, focused on superconductors. Its Neon model was trained on an open-weight base with experimental physics data and reached characterization performance comparable to OpenAI’s Astra — at roughly half the per-unit cost in internal testing, though that comparison flatters Neon (internal cost vs. retail inference price; assuming 50–70% OpenAI margins, the advantage disappears or inverts).
• OpenAI achieved parity with Periodic on characterization tasks despite having no domain-specific training data, in under a year — the key competitive vulnerability for science-only players is that frontier labs have too many other priorities to dedicate attention to any one domain, and attention is the only real moat.
• Three tiers of breakthrough: incremental (process efficiency compounds), novel (AI combines disparate approaches humans haven’t tried), fundamental (new physics — not yet observed, not required for the bull case). Incremental breakthroughs alone at 10x throughput look fundamental from the outside.
• Anthropic has opened a wet lab in San Francisco and is pursuing the full flywheel in biology, making it directly competitive with Lila — though the market is unbounded enough that Lila and Anthropic may never overlap in practice.

Notable quotes

“Much of tech looks at AI as labor replacement, which misses the exploration work that is being scaled up right now that will come to dwarf it. I’ve talked with dozens of founders, investors and scientists operating autonomous labs over several months — AI is now driving autonomous science and it’s both misunderstood and underestimated.”

“The constraint for science isn’t people or ideas, it’s verifying experiments. LLMs on their own can’t do the type of science that survives contact with the real world — they can read every paper and theorize down every rabbit hole but they can’t verify their hypotheses in a physical space.”

“AI enabling autonomous science, discovery, and exploration is the biggest bull case for AI — bottomless compute demand and next phases of AI build-out — that have precisely nothing to do with chatbots, personal agents, AI lawyers, or coding agents. Every other domain could see underwater economics, slow adoption, and profit skinning competition and it wouldn’t matter.”

“Breadth gives depth faster.” — Jon Hennek, CRO at Lila, on why cross-domain training outperforms narrow specialization

“Harvey v. Legora is competitive in the legal AI because neither is advancing new areas of law or developing novel theories. Lila v. Anthropic has no similarity to Harvey v. Legora whatsoever.”

“After spending time in the AI and science space, it’s a killshot for AI bearishness. While it’s easy to sound smart decrying AI-slop movies, the beAIux arts, and the end of middle-management, injecting raw intelligence into more science has almost pure upside.”

“Close your eyes and imagine warehouses full of lab stations operating 24/7 running tight experimentation loops with robotics. Then, smile, and think to yourself — with hundreds of thousands of PhDs in chemistry, biology, physics in the U.S. — how many stations could a PhD oversee? Could we produce 10x the science we do today?”

How it connects

• AI Agents — the enabling technology; autonomous lab operation is AI agents applied to physical science rather than digital tasks
• OpenAI — Astra set the initial capability baseline for lab automation; OpenAI also represents the frontier-lab threat to specialist players
• Anthropic — opened its own wet lab in San Francisco; the article names it as directly competitive with Lila in biology
• Flagship (Company) — Flagship Pioneering spun out Lila and previously spun out Moderna; the commercialization machine behind Lila’s GTM advantage
• Scientific Progress — the macro thesis: AI-autonomous science as the primary engine of technological progress in the next era
• Scientific Research — the institutional context being disrupted by autonomous labs
• Technological Innovation — the article frames autonomous scientific discovery as the most important AI application for long-run innovation