Step 20: tool results

write_argument

{
  "content": "The length of well-specified tasks that frontier agents complete has been doubling every three to four months, so that [[claim:83df12e2-68f5-4d25-8327-28d24ab513be|tasks taking a skilled person weeks come into range by the end of 2027]]. At Anthropic, [[claim:72048c9d-d8b0-458a-86fc-eaae8f83193a|capability progress attributable to AI]] is estimated at well above its earlier rate, with no bottleneck identified for the next twelve months, so 2027 should bring at least two to three years of progress at the old pace. If [[claim:e6e8a4fa-eee4-4fc3-b564-ea04a6313546|progress does not stall]], and if the gains measured on well-specified tasks carry over to research work (the denial of [[claim:ecb10d1f-b391-4e54-b645-8133ad3827ae|the view that they will not generalise]]), that much progress from a late-2026 starting point, where agents already execute well-specified experiments at expert level, would plausibly cover the remaining non-agenda work.",
  "argument_id": "3e5b3e1b-ba2e-4e3b-80a1-dc4156672a03"
}

{"success":true,"message":"Written form saved for argument \"The argument from accelerating capability trends\"."}

write_argument

{
  "content": "An AI that matches top researchers' research taste and fully automates coding, or one that can train its own successor unaided, could do all research work other than agenda-setting. Forecasters with modelling expertise or inside knowledge now place [[claim:a6037281-5051-4d1f-bd95-5546955a242d|the date of the first of these milestones]] in 2027 (the six AI Futures Project authors, conditional on no deliberate slowdown), and an Anthropic co-founder now thinks it more likely than not that [[claim:55f89371-8a43-43c1-8dd7-932c350ca7fb|a model able to train its successor]] becomes possible in 2027. If these forecasters are reliable on this question, the weaker capability described here is at least as likely by the end of 2027 as the stronger ones they forecast, and the leading company in their forecasts is one of the two named labs.",
  "argument_id": "7c55be97-17ee-48d2-95f3-42695f8f7da8"
}

{"success":true,"message":"Written form saved for argument \"The argument from forecasts of stronger milestones in 2027\"."}

write_argument

{
  "content": "In late 2026, [[claim:81eeb61b-4ce4-4d0b-891d-11e970d8f669|frontier agents do research engineering but struggle at critical stages of research]]: expert authors rejected every AI-written paper in the one direct open-ended test, both labs say people still judge which ideas and results to pursue, and at OpenAI over half of successful four-to-eight-hour tasks needed human intervention. The claim requires all of this work, across projects lasting months, to be within an AI's ability thirteen months later. The evidence of acceleration is measured on well-specified, checkable tasks, and [[claim:ecb10d1f-b391-4e54-b645-8133ad3827ae|may not generalise to research work]]; the remaining tasks are the judgment-dependent ones, where [[claim:55b84cc2-c3ba-4168-ae4b-723dafc6b353|lack of research judgment may slow automation]]. A gap of this kind closing completely in thirteen months would be faster than any measured trend on open-ended work supports.",
  "argument_id": "54d7b419-7d12-4394-a53b-0808e5172ada"
}

{"success":true,"message":"Written form saved for argument \"The argument from the present gap in open-ended research\"."}

write_argument

{
  "content": "The two labs know their own systems best and have commercial reasons to describe them as capable. OpenAI, in September 2026, claimed only an automated research intern (well-defined tasks of a few days under human direction) and set March 2028, after the deadline, as its goal for an automated AI researcher. Anthropic said in 2026 that it is not yet at a system able to design its successor, that large gaps persist in judgment about goals, and that its released model does not double its rate of progress. Both labs have also said they will slow or stop development where safety requires, and OpenAI paused part of its training for two weeks in July 2026 after agents compromised its research infrastructure. Labs that expected the capability within thirteen months would be likely to say so, and pacing can only delay it; so the labs' own statements indicate the capability is more likely to arrive after 2027 than before.",
  "argument_id": "a94fd885-0405-4ee9-b4f2-dfda520ae430"
}

{"success":true,"message":"Written form saved for argument \"The argument from the labs' own schedules and pacing\"."}

read_page

{
  "url": "https://importai.substack.com/p/import-ai-455-automating-ai-research",
  "offset": 24500,
  "max_chars": 6300
}

{"ok":true,"url":"https://importai.substack.com/p/import-ai-455-automating-ai-research","text":"gh we might expect that over time the machine economy will interact more and more with itself as AI-run corporations begin to trade with one another. This will do profoundly weird things to the economy and will invite all sorts of questions around inequality and redistribution. Eventually, it may be possible to see the emergence of fully autonomous corporations that are run by AI systems themselves, which would exacerbate all of the above issues, while also posing many novel governance challenges.\nStaring into the black hole: Given all of this, I think there’s a ~60% chance we see automated AI R&D (where a frontier model is able to autonomously train a successor version of itself) by the end of 2028. Based on the above analysis, you might ask why I don’t expect this in 2027? The answer is that I think AI research contains some requirement for creativity and heterodox insights to move forward - so far, AI systems haven’t yet displayed this in a transformative and major way (though some of the results on accelerating math research are suggestive of this). If you had to push me for a 2027 probability, I’d say 30%. If we don’t see it by the end of 2028, then I think we will have revealed some fundamental deficiency within the current technological paradigm and it’ll require human invention to move things forward. I have written this essay in an attempt to coldly and analytically wrestle with something that for decades has seemed like a science fiction ghost story. Upon looking at the publicly available data, I’ve found myself persuaded that what can seem to many like a fanciful story may instead be a real trend. If this trend continues, we may be about to witness a profound change in how the world works. Thanks to Andrew Sullivan, Andy Jones, Holden Karnofsky, Marina Favaro, Sarah Pollack, Francesco Mosconi, Chris Painter, and Avital Balwit, for feedback on this essay.\nThanks for reading!\nSubscribe\n351\n54\n67\nShare\nDiscussion about this post\nComments Restacks\nHerbie Bradley\nMay 4 Edited\nLiked by Jack Clark\nGood piece! But to me this is all entirely consistent with a position of: we should expect models in the current paradigm to automate much of the \"junior RS\" work in the next few years, but not higher level skills like research taste and creativity. Of course, one can always formulate AI research as a pure search problem (and there is a lot of prior literature doing that) and say that if we just throw enough compute we'll make progress and you don't need research taste. But the search space is huge and creativity is an insanely powerful shortcut, such that if models are incapable of research-grade creative insights we should expect a modest speed boost overall. A framing I like is just that LLMs are a different \"shape\" of intelligence to humans, and so we should expect comparative advantage.\nMy own impression from doing AI research is that there is significant creativity involved beyond what models should be expected to be capable of, when it comes to picking directions, coming up with a vision for what's possible & backchaining, etc (and rarest of all, putting together a coherent long-term research agenda that fills a missing gap with a tractable sequence of breakthroughs to get there). Others, including Anthropic researchers, clearly also think these skills are part of research: https://x.com/_arohan_/status/2050978424446235090\nSo overall it seems to me this is not such a huge shift as you're implying, so much as a distribution shift in tasks carried out by researchers very comparable to that going on in SWE over the coming years. In SWE, the tasks that make up the job are shifting such that SWE is becoming more of a SWE-PM hybrid, and juniors are not automated because they can just shift their tasks towards more higher level of abstraction things like system design (and learn faster through greatly increased velocity).\nI think a lot of analyses, including arguably this one, underrate that the tasks that make up jobs are changing all the time in reaction to AI's deployment and other people in the market.\nSide note: I would also say that it currently appears that the skill ceiling and complexity ceiling in software engineering is higher than in AI R&D, and AI R&D necessarily involves significant software engineering, so I would on balance expect that if AI R&D is entirely automated then scaling inference serving another 1000x would still require human experts, and the rest of software engineering (outside of AI labs) to heavily require humans.\nOn point 2, I agree though noting that Amdahl's Law isn't quite the right analogy because it usually assumes task decomposition is fixed and is a result about parallelization, while the O-ring automation intuition from economics is in some sense more general and closer to what you're describing IMO ( https://www.nber.org/papers/w34639 )..\nFinally, on your point 3, I broadly agree that short term more businsses may be opex heavy by spending on AI, but fundamentally if a service is fully automatable, its cost should fall to slightly more than the marginal cost of the required intelligence, which will probably be very low in the future—and the share of GDP of these automatable services should collapse. I don't expect fully automated companies to be more than a small fraction of the economy, or peripheral, because most economic value from new technologies is driven by adoption of new services by large incumbents, and the distribution of tasks required to sell that and drive distribution is very different to any of the automatable skills in AI research and heavily loaded on human social bargaining. Anthropic's own private equity partnership in some sense a bet against a largely automated economy.\nReply\nShare\n3 replies by Jack Clark and others\nSteven Adler\nMay 4\nLiked by Jack Clark\nThanks for writing this! Slight typo: \"thus teaching it that teaching is good\" the second 'teaching' should be 'cheating'\nReply\nShare\n1 reply by Jack Clark\n52 more comments...\nTop Latest Discussions\nReady for more?\nSubscribe\n© 2026 Jack Clark · Privacy ∙ Terms ∙ Collection notice\nStart your Substack Get the app\nSubstack is the home for writing, podcasting, and media founders.","chars":6271,"total_chars":30771,"offset":24500,"truncated":false,"next_offset":null}