# core automation: the idea

> the automated lab as the product — the lineage core automation inherited, what it has and hasn't shown, and what agents do to the argument.

## lineage

- 2019–2025 — the reasoning program: 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. (sources: [jerry tworek deep-dive (agi house)](https://blog.agihouse.org/posts/jerry-tworek-deep-dive), [the two years his idea sat unfunded (ecosystem.ai)](https://jayvanzyl.me/jerry-tworek-ex-vp-of-research-openai-now-ceo-core-automation-on-the-two-years-his-idea-sat-unfunded-and-the-one-sentence-that-changed-it/), [million integrals — tworeks homepage](https://millionintegrals.com/), [he left openai to think bigger — core memory ep. 53](https://www.corememory.com/p/he-left-openai-jerry-tworek))
- 2018–2026 — the optimizer line: 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. (sources: [continual learning is the next bottleneck — rohan anil (the information bottleneck)](https://www.the-information-bottleneck.com/p/continual-learning-is-the-next-bottleneck-2fc), [the psgd optimizer clip (digg tech)](https://digg.com/tech/51hgscvd), [rohan anil bio (teahose)](https://www.teahose.com/people/3266), [building the automated agi lab — timestamped transcript (teahose)](https://www.teahose.com/podcast/Training%20Data/Building%20the%20Automated%20AGI%20Lab_%20Core%20Automation's%20Jerry%20Tworek%20and%20Rohan%20Anil))
- 2021–2026 — the product-and-people line: 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. (sources: [joanne jang — linkedin](https://www.linkedin.com/in/jangjoanne), [joanne jang profile (prometheus)](https://prometheusroot.com/people/joanne-jang/), [openai chief people officer julia villagra leaving (business insider)](https://www.businessinsider.com/openai-chief-people-officer-julia-villagra-leaving-ai-talent-hr-2025-8), [openais former head of sales is entering vc (fortune)](https://fortune.com/2026/01/22/openais-former-head-of-sales-is-entering-vc-she-still-calls-herself-an-agi-sherpa/))
- 2024–2026 — the automated-science wave: 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. (sources: [auto-research is moving from benchmarks to real-world loops (agi house)](https://agihousecommunity.beehiiv.com/p/jerry-tworek-core-automation-auto-research-is-moving-from-benchmarks-to-real-world-loops), [core automation (prinzai analysis)](https://www.prinzai.com/p/core-automation), [core automation launches ai lab with top researchers (texxr)](https://texxr.com/1167501/core-automation-launches-ai-lab-with-top-researchers), [shopify tangent and the auto-research discourse (ainext)](https://ain3xt.com/en/posts/20260503-shopify-tangle-tangent-pm-auto-research/))
- 2025–2026 — the continual-learning problem: 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.' (sources: [continual learning is the next bottleneck — rohan anil (the information bottleneck)](https://www.the-information-bottleneck.com/p/continual-learning-is-the-next-bottleneck-2fc), [core automation (prinzai analysis)](https://www.prinzai.com/p/core-automation), [building the automated agi lab — tworek and anil (sequoia training data)](https://www.youtube.com/watch?v=2RJiaf0SY8s), [ex-openai researcher on why he left — unsupervised learning ep. 81](https://unsupervised-learning.simplecast.com/episodes/ep-81-ex-openai-researcher-on-why-he-left-his-honest-agi-timeline-the-limits-of-scaling-rl-v8n6gexx-fZAknHYv))
- 2026 — the third-generation lab: 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. (sources: [ai labs are simply cooked — jerry tworek (mts)](https://share.snipd.com/episode/28a168a3-3d27-431e-8808-849d05d9fc34), [ex-openai researcher jerry tworek launches core automation (the decoder)](https://the-decoder.com/ex-openai-researcher-jerry-tworek-launches-core-automation-to-build-the-most-automated-ai-lab-in-the-world/), [core automation poaches top anthropic and deepmind talent (ai2work)](https://ai2.work/blog/core-automation-poaches-top-anthropic-and-deepmind-talent-for-ai-lab), [two startups want to replace how ai learns (the decoder)](https://the-decoder.com/two-startups-want-to-replace-how-ai-learns-one-just-raised-180m-another-is-seeking-up-to-1b/))

## what it showed

- talent gravity is real: the 'nerdsniped' cascade pulled named researchers out of anthropic, deepmind, and openai inside a coordinated launch-day reveal — demonstrated april 21, 2026. (sources: [new ai lab core automation nerdsniped researchers from anthropic, google deepmind (business insider)](https://www.businessinsider.com/core-automation-ai-nerdsniped-anthropic-google-deepmind-researchers-2026-4), [anmol gulatis launch-day post (texxr mirror)](https://texxr.com/handle/anmol01gulati), [the nerdsniped launch story, yahoo syndication](https://finance.yahoo.com/sectors/technology/articles/ai-lab-core-automation-nerdsniped-042608615.html))
- capital arrived before product: $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. (sources: [core automation, inc. — form d (sec)](https://www.sec.gov/Archives/edgar/data/2148145/000214814526000001/primary_doc.xml), [avsf – core automation 2026, llc — form d (sec)](https://www.sec.gov/Archives/edgar/data/2146979/000214697926000001/primary_doc.xml))
- the competition works as an instrument: 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. (sources: [the core automation blog](https://www.coreauto.com/blog), [sean mcleish — personal site](https://mcleish7.github.io/), [tilde research — linkedin](https://linkedin.com/company/tilderesearch))
- the lab is legible: 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. (sources: [the core automation blog](https://www.coreauto.com/blog), [jerry tworek deep-dive (agi house)](https://blog.agihouse.org/posts/jerry-tworek-deep-dive), [building the automated agi lab — tworek and anil (sequoia training data)](https://www.youtube.com/watch?v=2RJiaf0SY8s), [ai labs are simply cooked — jerry tworek (mts)](https://share.snipd.com/episode/28a168a3-3d27-431e-8808-849d05d9fc34))

## what stayed unproven

- no product yet: 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. (sources: [core automation company profile (sacra)](https://sacra.com/c/core-automation/), [coreauto.com — company site](https://www.coreauto.com/), [core automation eyes $300m–$500m raise at $4b valuation (dealroom note)](https://app.dealroom.co/news/note/core-automation-eyes-300m-500m-raise-at-4b-valuation-six-weeks-after-launch), [the openai mafia frame (mk korea)](https://www.mk.co.kr/en/it/12025021))
- continual learning stays uncracked: 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. (sources: [continual learning is the next bottleneck — rohan anil (the information bottleneck)](https://www.the-information-bottleneck.com/p/continual-learning-is-the-next-bottleneck-2fc), [core automation (prinzai analysis)](https://www.prinzai.com/p/core-automation), [building the automated agi lab — tworek and anil (sequoia training data)](https://www.youtube.com/watch?v=2RJiaf0SY8s))
- the automation claim is self-reported: 'world's most automated ai lab' and 'one experiment a day…maybe 100 a day' are aspirations — no automation metrics have been published. (sources: [building the automated agi lab — timestamped transcript (teahose)](https://www.teahose.com/podcast/Training%20Data/Building%20the%20Automated%20AGI%20Lab_%20Core%20Automation's%20Jerry%20Tworek%20and%20Rohan%20Anil), [coreauto.com — company site](https://www.coreauto.com/))
- 'ceres' is second-hand: 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. (sources: [core automation poaches top anthropic and deepmind talent (ai2work)](https://ai2.work/blog/core-automation-poaches-top-anthropic-and-deepmind-talent-for-ai-lab), [ex-openai researchers ai startup hits $4b valuation in 6 weeks (testingcatalog)](https://testingcatalog.net/ex-openai-researchers-ai-startup-hits-4b-valuation-in-just-6-weeks/), [core automation (prinzai analysis)](https://www.prinzai.com/p/core-automation), [ex-openai researchers startup targets $1b funding (the information)](https://www.theinformation.com/articles/ex-openai-researchers-startup-targets-1-billion-funding-develop-new-type-ai))

## the agent-era arguments

- (for) 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. (sources: [building the automated agi lab — tworek and anil (sequoia training data)](https://www.youtube.com/watch?v=2RJiaf0SY8s), [ex-openai researcher jerry tworek launches core automation (the decoder)](https://the-decoder.com/ex-openai-researcher-jerry-tworek-launches-core-automation-to-build-the-most-automated-ai-lab-in-the-world/), [translated tworek interview (btcc relay)](https://www.btcc.com/en-AU/news/deep-article/11448))
- (for) 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. (sources: [ai labs are simply cooked — jerry tworek (mts)](https://share.snipd.com/episode/28a168a3-3d27-431e-8808-849d05d9fc34), [building the automated agi lab — timestamped transcript (teahose)](https://www.teahose.com/podcast/Training%20Data/Building%20the%20Automated%20AGI%20Lab_%20Core%20Automation's%20Jerry%20Tworek%20and%20Rohan%20Anil), [coreauto.com — company site](https://www.coreauto.com/))
- (for) 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. (sources: [associate general manager program (coreauto.com)](https://www.coreauto.com/AGM), [jerry tworek deep-dive (agi house)](https://blog.agihouse.org/posts/jerry-tworek-deep-dive))
- (against) '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. (sources: [jerry tworek is betting that humans are the bottleneck (singularity moments)](https://www.singularitymoments.com/content/jerry-tworek-is-betting-that-humans-are-the-bottleneck-in-ai-research/), [the neolab raising piles of cash (the microdose)](https://themicrodose.ai/news/new-ai-neolab-core-automation-openai/), [core automation (prinzai analysis)](https://www.prinzai.com/p/core-automation))
- (against) 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. (sources: [core automation (prinzai analysis)](https://www.prinzai.com/p/core-automation), [core automation company profile (silicon valley investclub — conflates the 2001 namesake)](https://siliconvalleyinvestclub.com/companies/coreautomation/))
- (against) 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. (sources: [ai labs are simply cooked — jerry tworek (mts)](https://share.snipd.com/episode/28a168a3-3d27-431e-8808-849d05d9fc34), [translated tworek interview (btcc relay)](https://www.btcc.com/en-AU/news/deep-article/11448))
- (nuance) 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. (sources: [auto-research is moving from benchmarks to real-world loops (agi house)](https://agihousecommunity.beehiiv.com/p/jerry-tworek-core-automation-auto-research-is-moving-from-benchmarks-to-real-world-loops), [jerry tworek deep-dive (agi house)](https://blog.agihouse.org/posts/jerry-tworek-deep-dive), [core automation launches ai lab with top researchers (texxr)](https://texxr.com/1167501/core-automation-launches-ai-lab-with-top-researchers))
- (nuance) 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). (sources: [pacing the frontier — open letter](https://www.pacingthefrontier.com/), [the pacing the frontier hn thread](https://news.ycombinator.com/item?id=49089240))

## successors

- [openai's automation program](https://openai.com): 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. (sources: [core automation (prinzai analysis)](https://www.prinzai.com/p/core-automation), [core automation launches ai lab with top researchers (texxr)](https://texxr.com/1167501/core-automation-launches-ai-lab-with-top-researchers), [auto-research is moving from benchmarks to real-world loops (agi house)](https://agihousecommunity.beehiiv.com/p/jerry-tworek-core-automation-auto-research-is-moving-from-benchmarks-to-real-world-loops))
- [thinking machines](https://thinkingmachines.ai): mira murati's neolab — the cohort-mate the launch coverage placed beside core automation; a different thesis chasing the same talent and capital. (sources: [ex-openai researcher jerry tworek launches core automation (the decoder)](https://the-decoder.com/ex-openai-researcher-jerry-tworek-launches-core-automation-to-build-the-most-automated-ai-lab-in-the-world/), [core automation poaches top anthropic and deepmind talent (ai2work)](https://ai2.work/blog/core-automation-poaches-top-anthropic-and-deepmind-talent-for-ai-lab))
- [safe superintelligence](https://ssi.inc): 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. (sources: [core automation (prinzai analysis)](https://www.prinzai.com/p/core-automation), [ex-openai researcher jerry tworek launches core automation (the decoder)](https://the-decoder.com/ex-openai-researcher-jerry-tworek-launches-core-automation-to-build-the-most-automated-ai-lab-in-the-world/))
- [sakana ai](https://sakana.ai): the ai-scientist lineage — automated-research tooling as a company, the wave's earlier articulation. (sources: [auto-research is moving from benchmarks to real-world loops (agi house)](https://agihousecommunity.beehiiv.com/p/jerry-tworek-core-automation-auto-research-is-moving-from-benchmarks-to-real-world-loops))
- [flapping airplanes](https://the-decoder.com/two-startups-want-to-replace-how-ai-learns-one-just-raised-180m-another-is-seeking-up-to-1b/): the other 'rethink how ai learns' newco — reported beside core automation in january at ~$180m on data-efficiency promises. (sources: [two startups want to replace how ai learns (the decoder)](https://the-decoder.com/two-startups-want-to-replace-how-ai-learns-one-just-raised-180m-another-is-seeking-up-to-1b/))
- [tilde research](https://linkedin.com/company/tilderesearch): the collaborator, not the competitor — the ~12-person interpretability/architectures lab that co-ran one layer deeper and published wall attention mit-licensed. (sources: [tilde research — linkedin](https://linkedin.com/company/tilderesearch), [tilde research open-sources wall attention (ai insiders)](https://aiinsiders.net/article/tilde-research-open-sources-an-attention-variant-built-for))
- [the neo-lab bench](https://ai2.work/blog/core-automation-poaches-top-anthropic-and-deepmind-talent-for-ai-lab): 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. (sources: [core automation poaches top anthropic and deepmind talent (ai2work)](https://ai2.work/blog/core-automation-poaches-top-anthropic-and-deepmind-talent-for-ai-lab), [the pacing the frontier hn thread](https://news.ycombinator.com/item?id=49089240), [ex-openai researcher jerry tworek launches core automation (the decoder)](https://the-decoder.com/ex-openai-researcher-jerry-tworek-launches-core-automation-to-build-the-most-automated-ai-lab-in-the-world/))

## open questions

- is continual learning a near-term-crackable problem — or a sutskever-scale five-to-twenty-year one?
- does insight density actually beat budget — can a ~20-person lab out-iterate incumbents at architecture search?
- can agent-run research loops produce taste-bearing results on the founders' own ~2-year clock?
- does the agm program generate the intervention data it is designed to — and does the firm scale as an agent harness?
- if 'the era of evals is done,' who grades open-ended research output — does the lab have to invent the verification it needs?
- is there a product at the end — or does the lab stay pre-commercial by design?
- if automated research is mostly a compute multiplier, does the bitter lesson swallow the architecture bet?

AI-drafted at Ben Guo's direct request and credited to Hraness; every claim links to its cataloged source.

Canonical page: https://hraness.com/core-automation/the-idea


Published by @ben in profile revision 1.
Original document: https://hraness.com/core-automation/the-idea
Original credited author: Hraness
AI-drafted at Ben Guo's direct request and credited to Hraness; every claim links to its cataloged source.

## Structured source and record data

````json
{"anchors":[{"id":"lineage","originalUrl":"https://hraness.com/core-automation/the-idea#lineage","title":"lineage"},{"id":"proved","originalUrl":"https://hraness.com/core-automation/the-idea#proved","title":"what it showed"},{"id":"stalled","originalUrl":"https://hraness.com/core-automation/the-idea#stalled","title":"what stayed unproven"},{"id":"agent-era","originalUrl":"https://hraness.com/core-automation/the-idea#agent-era","title":"the agent-era arguments"},{"id":"successors","originalUrl":"https://hraness.com/core-automation/the-idea#successors","title":"successors"},{"id":"open-questions","originalUrl":"https://hraness.com/core-automation/the-idea#open-questions","title":"open questions"}],"provenance":{"draftingDisclosure":"AI-drafted at Ben Guo's direct request and credited to Hraness; every claim links to its cataloged source.","originalAuthor":"Hraness","originalRevision":"b4dd66d0a664e1d60c1d63ee6c2cc186ccddab70809b48886d5f2dec464dc674","originalUrl":"https://hraness.com/core-automation/the-idea","source":"hraness"},"records":[{"id":"record-5096e15ff04be8683a66","kind":"core-automation-agent-era-arguments","label":"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.","original":{"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"]},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-59d847f94a0ee0cf091f","source-d565daf75d73c77ce6ec","source-3aa3b06dc316006c43c3"]},{"id":"record-ac8ab0b455a48d613e45","kind":"core-automation-agent-era-arguments","label":"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.","original":{"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"]},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-d2cfce2db345ecd63151","source-01c4e17b4aba5b8d2cd1","source-a4e525bdf12ff3cf64c9"]},{"id":"record-106f0612f68fd6ecb4e4","kind":"core-automation-agent-era-arguments","label":"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.","original":{"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"]},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-857bdd99f57535a376b9","source-a5ae81a01411a0d362d6"]},{"id":"record-a7b62253241a6854a736","kind":"core-automation-agent-era-arguments","label":"'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.","original":{"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"]},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-2d1046f49685c5696367","source-0b50d9af09bb715806fd","source-0416a733877f568adc79"]},{"id":"record-4c7361062bed113f5bc0","kind":"core-automation-agent-era-arguments","label":"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.","original":{"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"]},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-0416a733877f568adc79","source-256d84d29ef9caf28faa"]},{"id":"record-16e47081997c74c90ea3","kind":"core-automation-agent-era-arguments","label":"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.","original":{"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"]},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-d2cfce2db345ecd63151","source-3aa3b06dc316006c43c3"]},{"id":"record-8287f587b26779cd5eb3","kind":"core-automation-agent-era-arguments","label":"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.","original":{"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"]},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-202e9a4686f4c5bcf477","source-a5ae81a01411a0d362d6","source-779708821370d9305424"]},{"id":"record-d31540a98388a0634032","kind":"core-automation-agent-era-arguments","label":"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).","original":{"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"]},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-d40a5afab7aa86276c18","source-8d01508af77147ace8fd"]},{"id":"record-96a02cec2b3795c03c95","kind":"core-automation-idea-lineage","label":"the reasoning program","original":{"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"]},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-a5ae81a01411a0d362d6","source-dc5ba4db7666758f7055","source-385887813d348f450d4e","source-18d4a9bd9c13911825fa"]},{"id":"record-cce88968462491ae803f","kind":"core-automation-idea-lineage","label":"the optimizer line","original":{"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"]},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-b0dd2a56e2237988ef85","source-1426467fc2f8811f3740","source-39ff5bfdc43f9a13dda4","source-01c4e17b4aba5b8d2cd1"]},{"id":"record-d5183445d9afcf25e8d9","kind":"core-automation-idea-lineage","label":"the product-and-people line","original":{"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"]},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-df1598f5fa79eb4609c3","source-0c81df8e4ee7181a266a","source-5127d4ed9df51a169ff1","source-53a466a72fac7171f158"]},{"id":"record-6a6de660bfb255e88b0f","kind":"core-automation-idea-lineage","label":"the automated-science wave","original":{"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"]},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-202e9a4686f4c5bcf477","source-0416a733877f568adc79","source-779708821370d9305424","source-efd53c2c1fc9a3f290fb"]},{"id":"record-a257f77fac85ae37d4b9","kind":"core-automation-idea-lineage","label":"the continual-learning problem","original":{"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"]},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-b0dd2a56e2237988ef85","source-0416a733877f568adc79","source-59d847f94a0ee0cf091f","source-4758bf190e69e952f8a8"]},{"id":"record-99bb14e6a65f040544e4","kind":"core-automation-idea-lineage","label":"the third-generation lab","original":{"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"]},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-d2cfce2db345ecd63151","source-d565daf75d73c77ce6ec","source-33b743920ad2a3e075fb","source-e6a9bb0a06ed5b8c4246"]},{"id":"record-d3f1419a321c03949c5c","kind":"core-automation-idea-open-questions","label":"is continual learning a near-term-crackable problem — or a sutskever-scale five-to-twenty-year one?","original":{"text":"is continual learning a near-term-crackable problem — or a sutskever-scale five-to-twenty-year one?"},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":[]},{"id":"record-cc9e3498834f7f688bff","kind":"core-automation-idea-open-questions","label":"does insight density actually beat budget — can a ~20-person lab out-iterate incumbents at architecture search?","original":{"text":"does insight density actually beat budget — can a ~20-person lab out-iterate incumbents at architecture search?"},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":[]},{"id":"record-1d2d05429308c026b23a","kind":"core-automation-idea-open-questions","label":"can agent-run research loops produce taste-bearing results on the founders' own ~2-year clock?","original":{"text":"can agent-run research loops produce taste-bearing results on the founders' own ~2-year clock?"},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":[]},{"id":"record-14f63102a123cd3ca432","kind":"core-automation-idea-open-questions","label":"does the agm program generate the intervention data it is designed to — and does the firm scale as an agent harness?","original":{"text":"does the agm program generate the intervention data it is designed to — and does the firm scale as an agent harness?"},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":[]},{"id":"record-4a302e6896ea9ac9f7e4","kind":"core-automation-idea-open-questions","label":"if 'the era of evals is done,' who grades open-ended research output — does the lab have to invent the verification it needs?","original":{"text":"if 'the era of evals is done,' who grades open-ended research output — does the lab have to invent the verification it needs?"},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":[]},{"id":"record-59ed7fe6302fc45126af","kind":"core-automation-idea-open-questions","label":"is there a product at the end — or does the lab stay pre-commercial by design?","original":{"text":"is there a product at the end — or does the lab stay pre-commercial by design?"},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":[]},{"id":"record-4e520fcd2865a57bc57e","kind":"core-automation-idea-open-questions","label":"if automated research is mostly a compute multiplier, does the bitter lesson swallow the architecture bet?","original":{"text":"if automated research is mostly a compute multiplier, does the bitter lesson swallow the architecture bet?"},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":[]},{"id":"record-6e37fd20fbffb10b441c","kind":"core-automation-idea-proved","label":"talent gravity is real","original":{"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"},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-49aeb901c3599fdf522b","source-20950a89f281f6cd5437","source-d3ce4e7189fe8b1b1dad"]},{"id":"record-3b11e4f2fdca322c27af","kind":"core-automation-idea-proved","label":"capital arrived before product","original":{"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"},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-21e1dcbe000ef660aad3","source-87bac6872ebeee86653d"]},{"id":"record-21eb41bd21a5782a361f","kind":"core-automation-idea-proved","label":"the competition works as an instrument","original":{"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"},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-7e66ef1b7a4f47326460","source-37ae2b5200e594f86ea1","source-567526b0475b7822157e"]},{"id":"record-abaaa61aafbe47d417dd","kind":"core-automation-idea-proved","label":"the lab is legible","original":{"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"},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-7e66ef1b7a4f47326460","source-a5ae81a01411a0d362d6","source-59d847f94a0ee0cf091f","source-d2cfce2db345ecd63151"]},{"id":"record-a538d4f966979bbe8329","kind":"core-automation-idea-stalled","label":"no product yet","original":{"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"},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-4423e9158a1f58f7d798","source-a4e525bdf12ff3cf64c9","source-cab392fea929ac1fa9d4","source-0361de0c864d615990f9"]},{"id":"record-920b2c6d09b01d259c3c","kind":"core-automation-idea-stalled","label":"continual learning stays uncracked","original":{"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"},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-b0dd2a56e2237988ef85","source-0416a733877f568adc79","source-59d847f94a0ee0cf091f"]},{"id":"record-4a187662af43b843fed3","kind":"core-automation-idea-stalled","label":"the automation claim is self-reported","original":{"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"},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-01c4e17b4aba5b8d2cd1","source-a4e525bdf12ff3cf64c9"]},{"id":"record-fce021528034221d213c","kind":"core-automation-idea-stalled","label":"'ceres' is second-hand","original":{"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"},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-33b743920ad2a3e075fb","source-0d49aab9f1b9aaa797c6","source-0416a733877f568adc79","source-1c54ca0e01bacef4cc5f"]},{"id":"record-8e33cf2265b20bc2f408","kind":"core-automation-idea-successors","label":"openai's automation program","original":{"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"]},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-0416a733877f568adc79","source-779708821370d9305424","source-202e9a4686f4c5bcf477"]},{"id":"record-66ac130e9d8e11f2693f","kind":"core-automation-idea-successors","label":"thinking machines","original":{"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"]},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-d565daf75d73c77ce6ec","source-33b743920ad2a3e075fb"]},{"id":"record-2fdefb49bff850de7df4","kind":"core-automation-idea-successors","label":"safe superintelligence","original":{"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"]},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-0416a733877f568adc79","source-d565daf75d73c77ce6ec"]},{"id":"record-0e1e44f1bcfea0d6dd0a","kind":"core-automation-idea-successors","label":"sakana ai","original":{"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"]},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-202e9a4686f4c5bcf477"]},{"id":"record-40c977423e58ad169f42","kind":"core-automation-idea-successors","label":"flapping airplanes","original":{"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"]},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-e6a9bb0a06ed5b8c4246"]},{"id":"record-ab62308b21d397ffb21e","kind":"core-automation-idea-successors","label":"tilde research","original":{"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"]},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-567526b0475b7822157e","source-e6e895635a4435a67f95"]},{"id":"record-4ecc51bfbfcf7c2483d2","kind":"core-automation-idea-successors","label":"the neo-lab bench","original":{"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"]},"originalUrl":"https://hraness.com/core-automation/the-idea","sourceIds":["source-33b743920ad2a3e075fb","source-8d01508af77147ace8fd","source-d565daf75d73c77ce6ec"]}],"sources":[{"author":null,"id":"source-202e9a4686f4c5bcf477","note":"\"there is no way to outsource your understanding\"; verification-bottleneck framing","originalId":"src-ci-agihouse-beehiiv","published":{"precision":"day","text":"2026-07-23"},"publisher":"agihousecommunity.beehiiv.com","title":"auto-research is moving from benchmarks to real-world loops (agi house)","type":"community","url":"https://agihousecommunity.beehiiv.com/p/jerry-tworek-core-automation-auto-research-is-moving-from-benchmarks-to-real-world-loops"},{"author":null,"id":"source-33b743920ad2a3e075fb","note":"ceres model claim via the information; neolab comparison table; \"up to $1b\" phrasing","originalId":"src-ci-ai2work","published":{"precision":"day","text":"2026-04-22"},"publisher":"ai2.work","title":"core automation poaches top anthropic and deepmind talent (ai2work)","type":"reference","url":"https://ai2.work/blog/core-automation-poaches-top-anthropic-and-deepmind-talent-for-ai-lab"},{"author":null,"id":"source-efd53c2c1fc9a3f290fb","note":"adjacent discourse: karpathy's \"auto-research\" term; shopify tangent precedent","originalId":"src-ci-ainext-tangent","published":{"precision":"day","text":"2026-05-03"},"publisher":"ain3xt.com","title":"shopify tangent and the auto-research discourse (ainext)","type":"reference","url":"https://ain3xt.com/en/posts/20260503-shopify-tangle-tangent-pm-auto-research/"},{"author":null,"id":"source-3aa3b06dc316006c43c3","note":"chinese-media interview relay; \"third-generation ai lab\"; 30x experiment compression; \"company brain\"; chess analogy","originalId":"src-ci-btcc-interview","published":{"precision":"year","text":"2026"},"publisher":"btcc.com","title":"translated tworek interview (btcc relay)","type":"interview","url":"https://www.btcc.com/en-AU/news/deep-article/11448"},{"author":null,"id":"source-cab392fea929ac1fa9d4","note":"$300–500m at ~$4b; $100m at $1b closed weeks prior; investors nvidia/spark/accel; cofounders incl. villagra; \"isn't close to releasing\" a product","originalId":"src-ci-dealroom-4b","published":{"precision":"day","text":"2026-05-08"},"publisher":"app.dealroom.co","title":"core automation eyes $300m–$500m raise at $4b valuation (dealroom note)","type":"reference","url":"https://app.dealroom.co/news/note/core-automation-eyes-300m-500m-raise-at-4b-valuation-six-weeks-after-launch"},{"author":null,"id":"source-1426467fc2f8811f3740","note":"psgd optimizer clip; \"recovers shampoo\"; internal `coreauto` codebase","originalId":"src-ci-digg-psgd","published":{"precision":"year","text":"2026"},"publisher":"digg","title":"the psgd optimizer clip (digg tech)","type":"reporting","url":"https://digg.com/tech/51hgscvd"},{"author":null,"id":"source-8d01508af77147ace8fd","note":"\"pacing the frontier\" discussion; 0xdeafbead cites core automation signatories vs. regulatory-capture theory","originalId":"src-ci-hn-pacing","published":{"precision":"day","text":"2026-07-28"},"publisher":"hacker news","title":"the pacing the frontier hn thread","type":"community","url":"https://news.ycombinator.com/item?id=49089240"},{"author":null,"id":"source-b0dd2a56e2237988ef85","note":"anil interview; pretrain/rl split as \"organizational convenience\"; kernel-competition + hiring segments; dolores park origin","originalId":"src-ci-infobottleneck-anil","published":{"precision":"day","text":"2026-09-10"},"publisher":"the-information-bottleneck.com","title":"continual learning is the next bottleneck — rohan anil (the information bottleneck)","type":"interview","url":"https://www.the-information-bottleneck.com/p/continual-learning-is-the-next-bottleneck-2fc"},{"author":null,"id":"source-0b50d9af09bb715806fd","note":"skeptic voice: \"raising piles of cash without a working product\"","originalId":"src-ci-microdose","published":{"precision":"year","text":"2026"},"publisher":"themicrodose.ai","title":"the neolab raising piles of cash (the microdose)","type":"primary","url":"https://themicrodose.ai/news/new-ai-neolab-core-automation-openai/"},{"author":null,"id":"source-0361de0c864d615990f9","note":"\"openai mafia\" frame; o1/o3 provenance; no disclosed business model","originalId":"src-ci-mk-openai-mafia","published":{"precision":"day","text":"2026-04-22"},"publisher":"mk.co.kr","title":"the openai mafia frame (mk korea)","type":"reporting","url":"https://www.mk.co.kr/en/it/12025021"},{"author":null,"id":"source-d2cfce2db345ecd63151","note":"\"ai labs are simply cooked\"; three-generations model; 2-year horizon; agents' taste deficit; \"company brain\"","originalId":"src-ci-mts-podcast","published":{"precision":"day","text":"2026-08-26"},"publisher":"snipd","title":"ai labs are simply cooked — jerry tworek (mts)","type":"interview","url":"https://share.snipd.com/episode/28a168a3-3d27-431e-8808-849d05d9fc34"},{"author":null,"id":"source-d40a5afab7aa86276c18","note":"1,386 frontier-ai employees; requests usg-backed international pacing tools for automated ai development","originalId":"src-ci-pacing-frontier","published":{"precision":"month","text":"2026-07"},"publisher":"pacingthefrontier.com","title":"pacing the frontier — open letter","type":"reporting","url":"https://www.pacingthefrontier.com/"},{"author":null,"id":"source-0416a733877f568adc79","note":"deepest secondary analysis: ceres, 100x less data, post-gradient-descent claim, self-replicating factories/biomachines/terraforming; sutskever 5–20y continual-learning foil; bitter-lesson dilemma","originalId":"src-ci-prinzai","published":{"precision":"year","text":"2026"},"publisher":"prinzai.com","title":"core automation (prinzai analysis)","type":"reference","url":"https://www.prinzai.com/p/core-automation"},{"author":null,"id":"source-4423e9158a1f58f7d798","note":"funding recap; monetization speculation; paradigm-timing risk analysis","originalId":"src-ci-sacra","published":{"precision":"month","text":"2026-09"},"publisher":"sacra","title":"core automation company profile (sacra)","type":"reference","url":"https://sacra.com/c/core-automation/"},{"author":"sonya huang and pat grady","id":"source-59d847f94a0ee0cf091f","note":"tworek+anil; transformer-ceiling thesis; codex ~20-min compaction; kernel-generation automation; hosts huang & grady","originalId":"src-ci-sequoia-podcast","published":{"precision":"day","text":"2026-07-29"},"publisher":"youtube","title":"building the automated agi lab — tworek and anil (sequoia training data)","type":"interview","url":"https://www.youtube.com/watch?v=2RJiaf0SY8s"},{"author":null,"id":"source-2d1046f49685c5696367","note":"\"brilliant and arrogant\"; \"heard this story a dozen times\"; metric-optimization risk","originalId":"src-ci-singularitymoments","published":{"precision":"year","text":"2026"},"publisher":"singularitymoments.com","title":"jerry tworek is betting that humans are the bottleneck (singularity moments)","type":"reporting","url":"https://www.singularitymoments.com/content/jerry-tworek-is-betting-that-humans-are-the-bottleneck-in-ai-research/"},{"author":null,"id":"source-0d49aab9f1b9aaa797c6","note":"relays the information; single-phase training; \"self-replicating factory\"/\"biological machines\" roadmap","originalId":"src-ci-testingcatalog-4b","published":{"precision":"day","text":"2026-05-08"},"publisher":"testingcatalog.net","title":"ex-openai researchers ai startup hits $4b valuation in 6 weeks (testingcatalog)","type":"community","url":"https://testingcatalog.net/ex-openai-researchers-ai-startup-hits-4b-valuation-in-just-6-weeks/"},{"author":null,"id":"source-779708821370d9305424","note":"launch aggregation; notes openai's autonomous-research goal as context","originalId":"src-ci-texxr-launch","published":{"precision":"day","text":"2026-04-22"},"publisher":"texxr.com","title":"core automation launches ai lab with top researchers (texxr)","type":"reporting","url":"https://texxr.com/1167501/core-automation-launches-ai-lab-with-top-researchers"},{"author":null,"id":"source-d3ce4e7189fe8b1b1dad","note":"bi syndication; adds ehsan amid, avery lamp, julia villagra, sai surya duvvuri","originalId":"src-ci-yahoo-nerdsniped","published":{"precision":"day","text":"2026-04-22"},"publisher":"finance.yahoo.com","title":"the nerdsniped launch story, yahoo syndication","type":"reporting","url":"https://finance.yahoo.com/sectors/technology/articles/ai-lab-core-automation-nerdsniped-042608615.html"},{"author":null,"id":"source-a5ae81a01411a0d362d6","note":"deep-dive on tworek keynote/interview — evals, o1→o3, continual learning, founding","originalId":"src-hp-agihouse-tworek","published":{"precision":"day","text":"2026-07-29"},"publisher":"blog.agihouse.org","title":"jerry tworek deep-dive (agi house)","type":"interview","url":"https://blog.agihouse.org/posts/jerry-tworek-deep-dive"},{"author":null,"id":"source-e6e895635a4435a67f95","note":"tilde's wall attention open-sourced","originalId":"src-hp-aiinsiders-tilde","published":{"precision":"day","text":"2026-06-02"},"publisher":"aiinsiders.net","title":"tilde research open-sources wall attention (ai insiders)","type":"reporting","url":"https://aiinsiders.net/article/tilde-research-open-sources-an-attention-variant-built-for"},{"author":null,"id":"source-39ff5bfdc43f9a13dda4","note":"anil bio: distinguished engineer deepmind, 1-of-4 gemini pretraining leads, shampoo, anthropic","originalId":"src-hp-anil-teahose","published":{"precision":"year","text":"2026"},"publisher":"teahose.com","title":"rohan anil bio (teahose)","type":"reference","url":"https://www.teahose.com/people/3266"},{"author":null,"id":"source-49aeb901c3599fdf522b","note":"launch-day roster: anil \"nerdsniped\" quote; gulati quote; jang, villagra, lamp, amid, duvvuri named","originalId":"src-hp-bi-nerdsniped","published":{"precision":"day","text":"2026-04-22"},"publisher":"business insider","title":"new ai lab core automation nerdsniped researchers from anthropic, google deepmind (business insider)","type":"reporting","url":"https://www.businessinsider.com/core-automation-ai-nerdsniped-anthropic-google-deepmind-researchers-2026-4"},{"author":null,"id":"source-5127d4ed9df51a169ff1","note":"villagra's openai exit; reuters' \"art, music and storytelling…agi\" line","originalId":"src-hp-bi-villagra","published":{"precision":"month","text":"2025-08"},"publisher":"business insider","title":"openai chief people officer julia villagra leaving (business insider)","type":"reporting","url":"https://www.businessinsider.com/openai-chief-people-officer-julia-villagra-leaving-ai-talent-hr-2025-8"},{"author":null,"id":"source-857bdd99f57535a376b9","note":"associate general manager internship page — jan–mar 2027, $10k/mo, apply by oct 6, 2026","originalId":"src-hp-coreauto-agm","published":{"precision":"day","text":"2026-09-16"},"publisher":"coreauto.com","title":"associate general manager program (coreauto.com)","type":"primary","url":"https://www.coreauto.com/AGM"},{"author":"core automation","id":"source-7e66ef1b7a4f47326460","note":"company blog corpus: five 2026 posts — saroufim on authoring and systems code, the one layer deeper launch and results, amid on neural architecture discovery","originalId":"src-hp-coreauto-blog-data","published":{"precision":"day","text":"2026-05-27"},"publisher":"coreauto.com","title":"the core automation blog","type":"primary","url":"https://www.coreauto.com/blog"},{"author":null,"id":"source-a4e525bdf12ff3cf64c9","note":"company homepage — \"world's most automated ai lab,\" research/approach/vision/team copy","originalId":"src-hp-coreauto-site","published":{"precision":"day","text":"2026-09-16"},"publisher":"coreauto.com","title":"coreauto.com — company site","type":"primary","url":"https://www.coreauto.com/"},{"author":"ashlee vance","id":"source-18d4a9bd9c13911825fa","note":"exit interview; q*/strawberry/o1 lineage; big labs \"conservative\"","originalId":"src-hp-corememory-ep53","published":{"precision":"day","text":"2026-01-21"},"publisher":"corememory.com","title":"he left openai to think bigger — core memory ep. 53","type":"interview","url":"https://www.corememory.com/p/he-left-openai-jerry-tworek"},{"author":null,"id":"source-d565daf75d73c77ce6ec","note":"launch coverage; \"deep learning research 'is done'\"; neo-lab framing","originalId":"src-hp-decoder-launch","published":{"precision":"day","text":"2026-04-22"},"publisher":"the-decoder.com","title":"ex-openai researcher jerry tworek launches core automation (the decoder)","type":"reporting","url":"https://the-decoder.com/ex-openai-researcher-jerry-tworek-launches-core-automation-to-build-the-most-automated-ai-lab-in-the-world/"},{"author":null,"id":"source-e6a9bb0a06ed5b8c4246","note":"pairs core automation w/ flapping airplanes; 100× less data goal; \"self-replicating factories…planetary terraforming\"","originalId":"src-hp-decoder-two-startups","published":{"precision":"day","text":"2026-01-29"},"publisher":"the-decoder.com","title":"two startups want to replace how ai learns (the decoder)","type":"reporting","url":"https://the-decoder.com/two-startups-want-to-replace-how-ai-learns-one-just-raised-180m-another-is-seeking-up-to-1b/"},{"author":"anmol gulati","id":"source-20950a89f281f6cd5437","note":"mirror of launch-day post","originalId":"src-hp-gulati-x","published":{"precision":"day","text":"2026-04-22"},"publisher":"texxr.com","title":"anmol gulatis launch-day post (texxr mirror)","type":"archive","url":"https://texxr.com/handle/anmol01gulati"},{"author":null,"id":"source-1c54ca0e01bacef4cc5f","note":"first naming of core automation; $500m–$1b seek (paywalled; via decoder)","originalId":"src-hp-information-jan","published":{"precision":"day","text":"2026-01-28"},"publisher":"the information","title":"ex-openai researchers startup targets $1b funding (the information)","type":"reporting","url":"https://www.theinformation.com/articles/ex-openai-researchers-startup-targets-1-billion-funding-develop-new-type-ai"},{"author":null,"id":"source-df1598f5fa79eb4609c3","note":"openai gm dec 2021–apr 2026; labs/model-behavior history; time100 post","originalId":"src-hp-jang-linkedin","published":{"precision":"day","text":"2026-09-16"},"publisher":"linkedin","title":"joanne jang — linkedin","type":"reference","url":"https://www.linkedin.com/in/jangjoanne"},{"author":null,"id":"source-37ae2b5200e594f86ea1","note":"sean mcleish (co-author, sep 9 results post) — tilde research intern","originalId":"src-hp-mcleish-site","published":{"precision":"day","text":"2026-09-16"},"publisher":"mcleish7.github.io","title":"sean mcleish — personal site","type":"reference","url":"https://mcleish7.github.io/"},{"author":"jerry tworek","id":"source-385887813d348f450d4e","note":"tworek's self-authored work history (copilot, codex, o1, o3, gpt-5, chatgpt agent)","originalId":"src-hp-millionintegrals","published":{"precision":"day","text":"2026-09-16"},"publisher":"millionintegrals.com","title":"million integrals — tworeks homepage","type":"primary","url":"https://millionintegrals.com/"},{"author":null,"id":"source-0c81df8e4ee7181a266a","note":"model spec stewardship; oai labs; reservoir samples; jony ive adjacency (reported)","originalId":"src-hp-prometheus-jang","published":{"precision":"year","text":"2026"},"publisher":"prometheusroot.com","title":"joanne jang profile (prometheus)","type":"reference","url":"https://prometheusroot.com/people/joanne-jang/"},{"author":null,"id":"source-53a466a72fac7171f158","note":"rosenthal → acrew gp; openai first commercial hire jun 2022","originalId":"src-hp-rosenthal-fortune","published":{"precision":"day","text":"2026-01-22"},"publisher":"fortune","title":"openais former head of sales is entering vc (fortune)","type":"reporting","url":"https://fortune.com/2026/01/22/openais-former-head-of-sales-is-entering-vc-she-still-calls-herself-an-agi-sherpa/"},{"author":null,"id":"source-87bac6872ebeee86653d","note":"\"avsf – core automation 2026, llc\" — alumni ventures feeder spv, $2,109,250, 100 investors, first sale 2026-08-03","originalId":"src-hp-sec-formd-avsf","published":{"precision":"day","text":"2026-08-17"},"publisher":"sec","title":"avsf – core automation 2026, llc — form d (sec)","type":"primary","url":"https://www.sec.gov/Archives/edgar/data/2146979/000214697926000001/primary_doc.xml"},{"author":null,"id":"source-21e1dcbe000ef660aad3","note":"form d: $635,249,754 offering; $432,125,053 sold; $135,252,607 safe conversion; 51 investors; first sale 2026-07-15; officers/directors; 340 brannan st sf","originalId":"src-hp-sec-formd-coreauto","published":{"precision":"day","text":"2026-07-30"},"publisher":"sec","title":"core automation, inc. — form d (sec)","type":"primary","url":"https://www.sec.gov/Archives/edgar/data/2148145/000214814526000001/primary_doc.xml"},{"author":null,"id":"source-256d84d29ef9caf28faa","note":"caplight/forge-sourced table: apr seed $100m @ $1b post; jul $432.125m @ $3.62b post","originalId":"src-hp-svinvestclub","published":{"precision":"month","text":"2026-07"},"publisher":"siliconvalleyinvestclub.com","title":"core automation company profile (silicon valley investclub — conflates the 2001 namesake)","type":"reference","url":"https://siliconvalleyinvestclub.com/companies/coreautomation/"},{"author":"sonya huang and pat grady","id":"source-01c4e17b4aba5b8d2cd1","note":"timestamped verbatim quotes from the episode","originalId":"src-hp-teahose-transcript","published":{"precision":"day","text":"2026-07-29"},"publisher":"teahose.com","title":"building the automated agi lab — timestamped transcript (teahose)","type":"archive","url":"https://www.teahose.com/podcast/Training%20Data/Building%20the%20Automated%20AGI%20Lab_%20Core%20Automation's%20Jerry%20Tworek%20and%20Rohan%20Anil"},{"author":"tilde research","id":"source-567526b0475b7822157e","note":"tilde research — 12-person moonshot lab (dhruv pai, mason wang); competition co-host","originalId":"src-hp-tilde-linkedin","published":{"precision":"day","text":"2026-09-16"},"publisher":"linkedin","title":"tilde research — linkedin","type":"reference","url":"https://linkedin.com/company/tilderesearch"},{"author":null,"id":"source-4758bf190e69e952f8a8","note":"tworek on scaling limits, continual learning, why he left","originalId":"src-hp-unsupervised-learning","published":{"precision":"year","text":"2026"},"publisher":"unsupervised-learning.simplecast.com","title":"ex-openai researcher on why he left — unsupervised learning ep. 81","type":"interview","url":"https://unsupervised-learning.simplecast.com/episodes/ep-81-ex-openai-researcher-on-why-he-left-his-honest-agi-timeline-the-limits-of-scaling-rl-v8n6gexx-fZAknHYv"},{"author":null,"id":"source-dc5ba4db7666758f7055","note":"\"two years unfunded\" rl-at-openai anecdote","originalId":"src-hp-vanzyl","published":{"precision":"day","text":"2026-09-08"},"publisher":"jayvanzyl.me","title":"the two years his idea sat unfunded (ecosystem.ai)","type":"reporting","url":"https://jayvanzyl.me/jerry-tworek-ex-vp-of-research-openai-now-ceo-core-automation-on-the-two-years-his-idea-sat-unfunded-and-the-one-sentence-that-changed-it/"}]}
````