{"slug": "will-ai-build-itself-in-just-2-years", "title": "Will AI build itself in just 2 years?", "summary": "OpenAI and Anthropic have suggested that within two years, AI agents could improve ChatGPT and Claude without human employees, a scenario called recursive self-improvement. Epoch AI senior researcher JS Denain said this could be equivalent to Anthropic having 100 to 1,000 times more researchers working at 10 to 100 times the speed. However, recent research indicates AI still struggles with open-ended tasks such as setting research directions, incorporating critiques, and managing budgets.", "body_md": "Both [OpenAI](https://officechai.com/ai/recursive-self-improvement-openai-shares-prompt-with-which-it-used-gpt-5-6-sol-to-train-gpt-5-6-luna/) and [Anthropic](https://importai.substack.com/p/import-ai-455-automating-ai-research) have suggested that, in two years or less, they may no longer depend on their employees to improve ChatGPT and Claude. Instead, the work will be done by AI agents themselves.\n\n“You can imagine this as being equivalent to Anthropic [today], except it has 100 or 1,000 times more researchers than the company actually has,” Epoch AI senior researcher JS Denain told me, working through the hypothetical scenario in a Substack Live with *The Argument *Wednesday. “And also, all those researchers are working at 10 times, 100 times the speed.”\n\nThat scenario, which AI researchers call recursive self-improvement, would not necessarily mean that humans [lose control](https://www.theargumentmag.com/p/the-biggest-issue-in-american-politics) of AI. That said, most of the scary, sci-fi-adjacent [scenarios](https://ai-2040.com/?choices=plan-c-root) do at least [start](https://www.forethought.org/research/three-types-of-intelligence-explosion) from this premise.\n\nSo, just how close *are* we to self-improving AI?\n\nOne way to try to answer the question is to look at the recent overall arc of AI progress: How [long and complex](https://epoch.ai/publications/mirrorcode-preliminary-results) are the tasks that AI completes? How many [resources](https://situational-awareness.ai/from-gpt-4-to-agi/) are dedicated to this project? In both of those cases, progress has grown exponentially, lending weight to [bold predictions](https://www.aifuturesmodel.com/) about what could happen if such trends continue.\n\nAnother thing you could do is look at what’s happening inside the labs: OpenAI recently had its latest model [optimize](https://x.com/jxnlco/status/2075624940636975329) a smaller model. Anthropic [reported](https://www.anthropic.com/institute/recursive-self-improvement) that it tested Claude’s decision-making at certain critical junctures in the AI research process. Successive models made better and better decisions until Claude Mythos Preview beat human researchers 64% of the time.\n\nSuccessfully outsourcing AI research decisions to Mythos would be, to some, an obvious sign that agents could lead research soon. With this evidence in mind, [aggressive](https://importai.substack.com/p/import-ai-455-automating-ai-research) predictions that we’ll have automated AI researchers in just two years do not sound quite so far-fetched.\n\nBut making key decisions isn’t the same as running a whole scientific research process. That involves novel thinking, critical judgment, and resource management.\n\nTo figure out whether AI is up to *these* tasks, Denain has [proposed](https://epochai.substack.com/p/toward-an-onet-for-ai-r-and-d) treating AI research the way [economists](https://www.theargumentmag.com/p/will-we-know-when-ai-is-taking-our) treat any other automatable job in the economy: Itemize all the tasks involved in the job, then figure out how well AI can do each of them specifically.\n\nIn other words, what does it actually look like to be a scientist who is building AI? As I put it in our conversation: “What is the equivalent of an AI researcher dropping a drop of solution like a chemist does to run their own experiment?”\n\nDenain and I tried to separate things AI currently does well in the research process from the things it can’t yet do well.\n\nIt’s common to say that AI succeeds at defined tasks but struggles at more open-ended ones. This discussion shows there are still layers of complexity below that. You have to ask what particular aspects of open-ended tasks AI struggles with: Can it set new research directions? Can it incorporate critiques? Can it manage a budget?\n\nRecent [research](https://www.normaltech.ai/p/ai-agents-cant-yet-do-open-ended) suggests that, for all three, the answer, so far, is “no.” This complicates the picture of steady progress coming out of the AI labs. Watch the interview to learn more about why.", "url": "https://wpnews.pro/news/will-ai-build-itself-in-just-2-years", "canonical_source": "https://www.theargumentmag.com/p/will-ai-build-itself-in-just-2-years", "published_at": "2026-08-06 17:01:12+00:00", "updated_at": "2026-08-09 09:44:30.395748+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-research", "ai-agents"], "entities": ["OpenAI", "Anthropic", "Epoch AI", "JS Denain", "ChatGPT", "Claude", "Claude Mythos Preview"], "alternates": {"html": "https://wpnews.pro/news/will-ai-build-itself-in-just-2-years", "markdown": "https://wpnews.pro/news/will-ai-build-itself-in-just-2-years.md", "text": "https://wpnews.pro/news/will-ai-build-itself-in-just-2-years.txt", "jsonld": "https://wpnews.pro/news/will-ai-build-itself-in-just-2-years.jsonld"}}