Published text

Structured research records

1. architectural exploration is contestable — big labs are competitively locked into the transformer and coding-agent race, so the exploration a frontier incumbent can't fund becomes a ~20-person lab's opening. insight density versus budget is the live test. · core-automation-agent-era-arguments

   Original location · source-59d847f94a0ee0cf091f · source-d565daf75d73c77ce6ec · source-3aa3b06dc316006c43c3

   Original structured fields

   {
     "claim": "architectural exploration is contestable — big labs are competitively locked into the transformer and coding-agent race, so the exploration a frontier incumbent can't fund becomes a ~20-person lab's opening. insight density versus budget is the live test.",
     "id": "arg-contestable",
     "side": "for",
     "sourceIds": [
       "src-ci-sequoia-podcast",
       "src-hp-decoder-launch",
       "src-ci-btcc-interview"
     ]
   }

2. the two halves reinforce — organizational automation supplies iteration speed, a post-transformer architecture supplies the ceiling; if continual learning lands, the flywheel feeds on its own exhaust. · core-automation-agent-era-arguments

   Original location · source-d2cfce2db345ecd63151 · source-01c4e17b4aba5b8d2cd1 · source-a4e525bdf12ff3cf64c9

   Original structured fields

   {
     "claim": "the two halves reinforce — organizational automation supplies iteration speed, a post-transformer architecture supplies the ceiling; if continual learning lands, the flywheel feeds on its own exhaust.",
     "id": "arg-flywheel",
     "side": "for",
     "sourceIds": [
       "src-ci-mts-podcast",
       "src-hp-teahose-transcript",
       "src-hp-coreauto-site"
     ]
   }

3. real-world loops beat evals — the agm program is a data-generation scheme: humans run real businesses on agent infrastructure so the company can instrument exactly where they intervene. · core-automation-agent-era-arguments

   Original location · source-857bdd99f57535a376b9 · source-a5ae81a01411a0d362d6

   Original structured fields

   {
     "claim": "real-world loops beat evals — the agm program is a data-generation scheme: humans run real businesses on agent infrastructure so the company can instrument exactly where they intervene.",
     "id": "arg-realworld",
     "side": "for",
     "sourceIds": [
       "src-hp-coreauto-agm",
       "src-hp-agihouse-tworek"
     ]
   }

4. 'old promise, new label' — automated discovery has been promised for a decade, 'if this tech were easy, the trillion-dollar labs would have built it yesterday,' and a self-driving lab optimizes whatever metric you hand it. · core-automation-agent-era-arguments

   Original location · source-2d1046f49685c5696367 · source-0b50d9af09bb715806fd · source-0416a733877f568adc79

   Original structured fields

   {
     "claim": "'old promise, new label' — automated discovery has been promised for a decade, 'if this tech were easy, the trillion-dollar labs would have built it yesterday,' and a self-driving lab optimizes whatever metric you hand it.",
     "id": "arg-old-promise",
     "side": "against",
     "sourceIds": [
       "src-ci-singularitymoments",
       "src-ci-microdose",
       "src-ci-prinzai"
     ]
   }

5. the bitter-lesson trap — if automated research is mostly a compute multiplier, the biggest compute holder wins; ~$532m is real money and small against the labs core automation claims to out-iterate. · core-automation-agent-era-arguments

   Original location · source-0416a733877f568adc79 · source-256d84d29ef9caf28faa

   Original structured fields

   {
     "claim": "the bitter-lesson trap — if automated research is mostly a compute multiplier, the biggest compute holder wins; ~$532m is real money and small against the labs core automation claims to out-iterate.",
     "id": "arg-bitter-lesson",
     "side": "against",
     "sourceIds": [
       "src-ci-prinzai",
       "src-hp-svinvestclub"
     ]
   }

6. the taste problem — the founders themselves concede today's agents 'lack deep field understanding and produce low-quality, high-creativity ideas.' the two-year horizon for agents doing research is an assertion, not a demonstrated curve. · core-automation-agent-era-arguments

   Original location · source-d2cfce2db345ecd63151 · source-3aa3b06dc316006c43c3

   Original structured fields

   {
     "claim": "the taste problem — the founders themselves concede today's agents 'lack deep field understanding and produce low-quality, high-creativity ideas.' the two-year horizon for agents doing research is an assertion, not a demonstrated curve.",
     "id": "arg-taste",
     "side": "against",
     "sourceIds": [
       "src-ci-mts-podcast",
       "src-ci-btcc-interview"
     ]
   }

7. the verification gap — 'the era of evals is done,' but no real-world verification loop exists for open-ended research output; the lab has to invent its own grading while racing its founder's old employer to the same destination. · core-automation-agent-era-arguments

   Original location · source-202e9a4686f4c5bcf477 · source-a5ae81a01411a0d362d6 · source-779708821370d9305424

   Original structured fields

   {
     "claim": "the verification gap — 'the era of evals is done,' but no real-world verification loop exists for open-ended research output; the lab has to invent its own grading while racing its founder's old employer to the same destination.",
     "id": "arg-verification",
     "side": "nuance",
     "sourceIds": [
       "src-ci-agihouse-beehiiv",
       "src-hp-agihouse-tworek",
       "src-ci-texxr-launch"
     ]
   }

8. the pacing wrinkle — the company automating frontier research signed the letter asking for tools to pace frontier research. readable as consistency (build governance with the accelerator) or tension (the accelerator asking for brakes). · core-automation-agent-era-arguments

   Original location · source-d40a5afab7aa86276c18 · source-8d01508af77147ace8fd

   Original structured fields

   {
     "claim": "the pacing wrinkle — the company automating frontier research signed the letter asking for tools to pace frontier research. readable as consistency (build governance with the accelerator) or tension (the accelerator asking for brakes).",
     "id": "arg-pacing",
     "side": "nuance",
     "sourceIds": [
       "src-ci-pacing-frontier",
       "src-ci-hn-pacing"
     ]
   }

9. the reasoning program · core-automation-idea-lineage

   Original location · source-a5ae81a01411a0d362d6 · source-dc5ba4db7666758f7055 · source-385887813d348f450d4e · source-18d4a9bd9c13911825fa

   Original structured fields

   {
     "contribution": "the lesson tworek exported from openai: the rl-reasoning bet sat unfunded for about two years inside the lab until 'now we have the gpus, try to scale it' — then produced o1, o3, and gpt-5's reasoning. big ideas need a patron with gpus; at core automation he is the patron.",
     "era": "2019–2025",
     "id": "lin-reasoning",
     "name": "the reasoning program",
     "sourceIds": [
       "src-hp-agihouse-tworek",
       "src-hp-vanzyl",
       "src-hp-millionintegrals",
       "src-hp-corememory-ep53"
     ]
   }

10. the optimizer line · core-automation-idea-lineage

    Original location · source-b0dd2a56e2237988ef85 · source-1426467fc2f8811f3740 · source-39ff5bfdc43f9a13dda4 · source-01c4e17b4aba5b8d2cd1

    Original structured fields

    {
      "contribution": "anil's stack — google brain systems engineering, the shampoo optimizer, palm/gemini pre-training optimization, anthropic pre-training — arriving at the psgd framework implemented inside the internal coreauto codebase, with the ~60× qr-factorization kernel result as its public proof point.",
      "era": "2018–2026",
      "id": "lin-optimizer",
      "name": "the optimizer line",
      "sourceIds": [
        "src-ci-infobottleneck-anil",
        "src-ci-digg-psgd",
        "src-hp-anil-teahose",
        "src-hp-teahose-transcript"
      ]
    }

11. the product-and-people line · core-automation-idea-lineage

    Original location · source-df1598f5fa79eb4609c3 · source-0c81df8e4ee7181a266a · source-5127d4ed9df51a169ff1 · source-53a466a72fac7171f158

    Original structured fields

    {
      "contribution": "the second half of the stack — jang's model behavior and model spec work, villagra's people-scaling through openai's hypergrowth, rosenthal's go-to-market record. a lab that automates work needs behavior, org, and distribution instincts, not just models.",
      "era": "2021–2026",
      "id": "lin-product-people",
      "name": "the product-and-people line",
      "sourceIds": [
        "src-hp-jang-linkedin",
        "src-hp-prometheus-jang",
        "src-hp-bi-villagra",
        "src-hp-rosenthal-fortune"
      ]
    }

12. the automated-science wave · core-automation-idea-lineage

    Original location · source-202e9a4686f4c5bcf477 · source-0416a733877f568adc79 · source-779708821370d9305424 · source-efd53c2c1fc9a3f290fb

    Original structured fields

    {
      "contribution": "the ambient wave the lab rides — sakana's ai scientist lineage, karpathy-coined 'auto-research' discourse, shopify's 'tangent' precedent for agent-run internal projects, and openai's own stated goal of fully automated ai research by march 2028, which makes the founder's old employer the explicit competitor.",
      "era": "2024–2026",
      "id": "lin-autoresearch",
      "name": "the automated-science wave",
      "sourceIds": [
        "src-ci-agihouse-beehiiv",
        "src-ci-prinzai",
        "src-ci-texxr-launch",
        "src-ci-ainext-tangent"
      ]
    }

13. the continual-learning problem · core-automation-idea-lineage

    Original location · source-b0dd2a56e2237988ef85 · source-0416a733877f568adc79 · source-59d847f94a0ee0cf091f · source-4758bf190e69e952f8a8

    Original structured fields

    {
      "contribution": "the unsolved problem the whole bet rests on — models that keep learning after deployment, blocked by catastrophic forgetting. sutskever's reported 5–20-year estimate is the foil: the founders argue it's nearer-term crackable, and the pre-training/rl split is 'organizational convenience rather than science.'",
      "era": "2025–2026",
      "id": "lin-continual",
      "name": "the continual-learning problem",
      "sourceIds": [
        "src-ci-infobottleneck-anil",
        "src-ci-prinzai",
        "src-ci-sequoia-podcast",
        "src-hp-unsupervised-learning"
      ]
    }

14. the third-generation lab · core-automation-idea-lineage

    Original location · source-d2cfce2db345ecd63151 · source-d565daf75d73c77ce6ec · source-33b743920ad2a3e075fb · source-e6a9bb0a06ed5b8c4246

    Original structured fields

    {
      "contribution": "tworek's three-generation frame — gen-1 research-driven labs (early deepmind, early openai), gen-2 compute-scaling labs (post-gpt-3), gen-3 automation-first labs. core automation claims the third generation: ~20 people, agents doing the experiments, insight density over budget.",
      "era": "2026",
      "id": "lin-neolab",
      "name": "the third-generation lab",
      "sourceIds": [
        "src-ci-mts-podcast",
        "src-hp-decoder-launch",
        "src-ci-ai2work",
        "src-hp-decoder-two-startups"
      ]
    }

15. is continual learning a near-term-crackable problem — or a sutskever-scale five-to-twenty-year one? · core-automation-idea-open-questions

    Original location

    Original structured fields

    {
      "text": "is continual learning a near-term-crackable problem — or a sutskever-scale five-to-twenty-year one?"
    }

16. does insight density actually beat budget — can a ~20-person lab out-iterate incumbents at architecture search? · core-automation-idea-open-questions

    Original location

    Original structured fields

    {
      "text": "does insight density actually beat budget — can a ~20-person lab out-iterate incumbents at architecture search?"
    }

17. can agent-run research loops produce taste-bearing results on the founders' own ~2-year clock? · core-automation-idea-open-questions

    Original location

    Original structured fields

    {
      "text": "can agent-run research loops produce taste-bearing results on the founders' own ~2-year clock?"
    }

18. does the agm program generate the intervention data it is designed to — and does the firm scale as an agent harness? · core-automation-idea-open-questions

    Original location

    Original structured fields

    {
      "text": "does the agm program generate the intervention data it is designed to — and does the firm scale as an agent harness?"
    }

19. if 'the era of evals is done,' who grades open-ended research output — does the lab have to invent the verification it needs? · core-automation-idea-open-questions

    Original location

    Original structured fields

    {
      "text": "if 'the era of evals is done,' who grades open-ended research output — does the lab have to invent the verification it needs?"
    }

20. is there a product at the end — or does the lab stay pre-commercial by design? · core-automation-idea-open-questions

    Original location

    Original structured fields

    {
      "text": "is there a product at the end — or does the lab stay pre-commercial by design?"
    }

21. if automated research is mostly a compute multiplier, does the bitter lesson swallow the architecture bet? · core-automation-idea-open-questions

    Original location

    Original structured fields

    {
      "text": "if automated research is mostly a compute multiplier, does the bitter lesson swallow the architecture bet?"
    }

22. talent gravity is real · core-automation-idea-proved

    Original location · source-49aeb901c3599fdf522b · source-20950a89f281f6cd5437 · source-d3ce4e7189fe8b1b1dad

    Original structured fields

    {
      "detail": "the 'nerdsniped' cascade pulled named researchers out of anthropic, deepmind, and openai inside a coordinated launch-day reveal — demonstrated april 21, 2026.",
      "id": "clm-talent-gravity",
      "sourceIds": [
        "src-hp-bi-nerdsniped",
        "src-hp-gulati-x",
        "src-ci-yahoo-nerdsniped"
      ],
      "title": "talent gravity is real"
    }

23. capital arrived before product · core-automation-idea-proved

    Original location · source-21e1dcbe000ef660aad3 · source-87bac6872ebeee86653d

    Original structured fields

    {
      "detail": "$432,125,053 actually sold to 51 investors per the july 30 form d, plus a $2.1m alumni ventures feeder spv with its own filing — demand confirmed even where the valuation stays reported.",
      "id": "clm-capital-first",
      "sourceIds": [
        "src-hp-sec-formd-coreauto",
        "src-hp-sec-formd-avsf"
      ],
      "title": "capital arrived before product"
    }

24. the competition works as an instrument · core-automation-idea-proved

    Original location · source-7e66ef1b7a4f47326460 · source-37ae2b5200e594f86ea1 · source-567526b0475b7822157e

    Original structured fields

    {
      "detail": "one layer deeper produced a real measurement: 15,602 submissions, 18 certified hard uploads from 5 accounts, winners wiring around python primitives. test-time depth is genuinely hard for current methods — measured september 9, 2026.",
      "id": "clm-competition",
      "sourceIds": [
        "src-hp-coreauto-blog-data",
        "src-hp-mcleish-site",
        "src-hp-tilde-linkedin"
      ],
      "title": "the competition works as an instrument"
    }

25. the lab is legible · core-automation-idea-proved

    Original location · source-7e66ef1b7a4f47326460 · source-a5ae81a01411a0d362d6 · source-59d847f94a0ee0cf091f · source-d2cfce2db345ecd63151

    Original structured fields

    {
      "detail": "five months produced five blog posts, a keynote with an open-problems slide eighteen teams built on, and flagship podcast episodes — the thesis got a fair technical hearing in public.",
      "id": "clm-public-lab",
      "sourceIds": [
        "src-hp-coreauto-blog-data",
        "src-hp-agihouse-tworek",
        "src-ci-sequoia-podcast",
        "src-ci-mts-podcast"
      ],
      "title": "the lab is legible"
    }

26. no product yet · core-automation-idea-stalled

    Original location · source-4423e9158a1f58f7d798 · source-a4e525bdf12ff3cf64c9 · source-cab392fea929ac1fa9d4 · source-0361de0c864d615990f9

    Original structured fields

    {
      "detail": "no api, no pricing, no papers, no public benchmarks as of september 2026 — pre-revenue by design; the near-term 'product' is the lab's own research loop.",
      "id": "clm-no-product",
      "sourceIds": [
        "src-ci-sacra",
        "src-hp-coreauto-site",
        "src-ci-dealroom-4b",
        "src-ci-mk-openai-mafia"
      ],
      "title": "no product yet"
    }

27. continual learning stays uncracked · core-automation-idea-stalled

    Original location · source-b0dd2a56e2237988ef85 · source-0416a733877f568adc79 · source-59d847f94a0ee0cf091f

    Original structured fields

    {
      "detail": "no public evidence of the lab's approach to continual learning — the bet is that a post-transformer architecture exists and a ~20-person team finds it first.",
      "id": "clm-continual-unproven",
      "sourceIds": [
        "src-ci-infobottleneck-anil",
        "src-ci-prinzai",
        "src-ci-sequoia-podcast"
      ],
      "title": "continual learning stays uncracked"
    }

28. the automation claim is self-reported · core-automation-idea-stalled

    Original location · source-01c4e17b4aba5b8d2cd1 · source-a4e525bdf12ff3cf64c9

    Original structured fields

    {
      "detail": "'world's most automated ai lab' and 'one experiment a day…maybe 100 a day' are aspirations — no automation metrics have been published.",
      "id": "clm-automation-claim",
      "sourceIds": [
        "src-hp-teahose-transcript",
        "src-hp-coreauto-site"
      ],
      "title": "the automation claim is self-reported"
    }

29. 'ceres' is second-hand · core-automation-idea-stalled

    Original location · source-33b743920ad2a3e075fb · source-0d49aab9f1b9aaa797c6 · source-0416a733877f568adc79 · source-1c54ca0e01bacef4cc5f

    Original structured fields

    {
      "detail": "the reported flagship model — single-phase, ~100× less data, continuous deployment learning — reached the record through the information's relays and was never confirmed by the company.",
      "id": "clm-ceres",
      "sourceIds": [
        "src-ci-ai2work",
        "src-ci-testingcatalog-4b",
        "src-ci-prinzai",
        "src-hp-information-jan"
      ],
      "title": "'ceres' is second-hand"
    }

30. openai's automation program · core-automation-idea-successors

    Original location · source-0416a733877f568adc79 · source-779708821370d9305424 · source-202e9a4686f4c5bcf477

    Original structured fields

    {
      "href": "https://openai.com",
      "id": "suc-openai",
      "name": "openai's automation program",
      "note": "the explicit foil — openai's internal program reportedly targets fully automated ai research by march 2028; anthropic and google deepmind run parallel internal automation programs. the founder's old employer is the benchmark.",
      "sourceIds": [
        "src-ci-prinzai",
        "src-ci-texxr-launch",
        "src-ci-agihouse-beehiiv"
      ]
    }

31. thinking machines · core-automation-idea-successors

    Original location · source-d565daf75d73c77ce6ec · source-33b743920ad2a3e075fb

    Original structured fields

    {
      "href": "https://thinkingmachines.ai",
      "id": "suc-thinking-machines",
      "name": "thinking machines",
      "note": "mira murati's neolab — the cohort-mate the launch coverage placed beside core automation; a different thesis chasing the same talent and capital.",
      "sourceIds": [
        "src-hp-decoder-launch",
        "src-ci-ai2work"
      ]
    }

32. safe superintelligence · core-automation-idea-successors

    Original location · source-0416a733877f568adc79 · source-d565daf75d73c77ce6ec

    Original structured fields

    {
      "href": "https://ssi.inc",
      "id": "suc-ssi",
      "name": "safe superintelligence",
      "note": "ilya sutskever's lab — and the named foil on continual learning, where his reported 5–20-year estimate sits against the founders' nearer-term bet.",
      "sourceIds": [
        "src-ci-prinzai",
        "src-hp-decoder-launch"
      ]
    }

33. sakana ai · core-automation-idea-successors

    Original location · source-202e9a4686f4c5bcf477

    Original structured fields

    {
      "href": "https://sakana.ai",
      "id": "suc-sakana",
      "name": "sakana ai",
      "note": "the ai-scientist lineage — automated-research tooling as a company, the wave's earlier articulation.",
      "sourceIds": [
        "src-ci-agihouse-beehiiv"
      ]
    }

34. flapping airplanes · core-automation-idea-successors

    Original location · source-e6a9bb0a06ed5b8c4246

    Original structured fields

    {
      "href": "https://the-decoder.com/two-startups-want-to-replace-how-ai-learns-one-just-raised-180m-another-is-seeking-up-to-1b/",
      "id": "suc-flapping-airplanes",
      "name": "flapping airplanes",
      "note": "the other 'rethink how ai learns' newco — reported beside core automation in january at ~$180m on data-efficiency promises.",
      "sourceIds": [
        "src-hp-decoder-two-startups"
      ]
    }

35. tilde research · core-automation-idea-successors

    Original location · source-567526b0475b7822157e · source-e6e895635a4435a67f95

    Original structured fields

    {
      "href": "https://linkedin.com/company/tilderesearch",
      "id": "suc-tilde",
      "name": "tilde research",
      "note": "the collaborator, not the competitor — the ~12-person interpretability/architectures lab that co-ran one layer deeper and published wall attention mit-licensed.",
      "sourceIds": [
        "src-hp-tilde-linkedin",
        "src-hp-aiinsiders-tilde"
      ]
    }

36. the neo-lab bench · core-automation-idea-successors

    Original location · source-33b743920ad2a3e075fb · source-8d01508af77147ace8fd · source-d565daf75d73c77ce6ec

    Original structured fields

    {
      "href": "https://ai2.work/blog/core-automation-poaches-top-anthropic-and-deepmind-talent-for-ai-lab",
      "id": "suc-neolab-cohort",
      "name": "the neo-lab bench",
      "note": "the rest of the wave — periodic labs, ami labs (lecun's world-model lab), adaption labs (hooker), inherent, prime intellect — competing for the same ~1,000-person frontier-talent pool and the same neolab capital.",
      "sourceIds": [
        "src-ci-ai2work",
        "src-ci-hn-pacing",
        "src-hp-decoder-launch"
      ]
    }