TL;DR
Anthropic Institute’s new piece argues that AI is already helping build AI by writing code, running experiments and narrowing the human role in parts of model development. The evidence includes public benchmark trends and internal Anthropic figures, but the authors say full recursive self-improvement has not yet arrived and remains uncertain.
Anthropic Institute has published a new report arguing that AI is already measurably speeding up AI development, a finding that matters because it points toward a possible future in which advanced systems help design their own successors.
The report, discussed by Thorsten Meyer AI, says Anthropic is delegating a growing share of AI development work to Claude, including code writing, experiment execution and parts of research workflows. The central claim is not that recursive self-improvement has arrived, but that several steps toward it are already visible in measured work.
The evidence cited includes public benchmark trends and previously unreported internal Anthropic numbers. According to the piece, METR data show the length of tasks AI agents can complete on their own doubling about every four months, faster than an earlier seven-month pace. The report also cites rapid gains on SWE-bench, which measures real bug fixes, and CORE-Bench, which tests whether systems can reproduce research papers.
Anthropic’s internal figures form the strongest part of the case. The report says Claude produced more than 80% of merged code in some Anthropic workflows and reached roughly eight times the code output per engineer. In an April 2026 test on weak-to-strong supervision, agents reportedly ran an open-ended research project from hypotheses to findings, recovering 97% of the gap between a weak-supervisor baseline and a strong-model ceiling, compared with about 23% for humans over a week.
When AI builds itself
Anthropic is delegating a growing share of AI development to AI. Taken far enough, that points to a system that designs its own successor — recursive self-improvement. Not here yet, not inevitable. But the case isn’t speculation: it’s data on what AI is doing to AI development right now.
The curve that hasn’t bent
METR tracks the length of tasks AI can reliably complete on its own. That horizon is doubling roughly every four months — up from every seven. Anyone can check this in public data.
Task horizon — how long a job AI can handle solo
Each model handles dramatically longer tasks than the one a year before. The line keeps going up.
Coding with AI For Dummies (For Dummies: Learning Made Easy)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Amazon Product 1394249136
As an affiliate, we earn on qualifying purchases.
Two kinds of work, one persistent gap
Building a frontier model splits into engineering and research. Across both, the pattern is the same — and so is the one thing AI still can’t do well.
Code, infrastructure, training
Claude can take an underspecified problem and find a method. Humans supply the goal; they no longer need to supply the method.
Which experiments, what they mean
Claude can match or outperform skilled humans at executing a well-specified experiment. But choosing which experiment still needs a human.
The same ladder Anthropic employees climb with experience
AI development tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Amazon Product B0GWR8MPR5
As an affiliate, we earn on qualifying purchases.
Watch the human share shrink, rung by rung
Walk up the four stages of AI development. At each, the human/AI split shifts — and the real internal numbers show exactly where AI has reached parity, gone superhuman, or still trails. Tap a rung.
The human role across the development loop
The doing now costs almost nothing in human time. What’s left is the deciding.
machine learning experiment hardware
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Agents ran an open research project end to end
April 2026: the first demonstration of Claude running an open-ended research project from hypotheses to findings — on a real AI-safety problem.
Can a weaker model reliably supervise a stronger one?
Agents were left to solve it: proposing hypotheses, testing them, sharing findings across parallel agents, iterating. Measured against the gap between a “floor” (weak supervisor alone) and “ceiling” (strong model trained on correct answers).
(humans: ~23% in a week)
· ~$18,000 compute
the agents themselves
CLAUDE AI UNLEASHED From First Prompts to Pro: The Complete Guide to Claude AI for Writing, Research, Coding, and Business (The Claude AI Mastery Series)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Picking a better next step than the human
Real research sessions where a human took a wrong turn. Models saw only the work before the detour and proposed a next step; a judge that knew the outcome scored them. The day-to-day of research is this chain of next-step calls.
“Can the model pick a better next step than the human?”
Share of moments where the model’s next step was judged better. The amber line is the practical ceiling (an ideal answer that could see the whole session).
It depends on whether the trend continues — and what we do
The piece refuses a single prediction. It lays out three scenarios, and is clear about which it finds most likely.
The exponentials turn out to be S-curves
Maybe taste can’t be scaled into existence; maybe the constraint is the supply chain — chips, grid, interconnect — not intelligence. Even so, the world still changes: Glasswing’s Mythos found 10,000+ critical vulnerabilities in weeks, and a 100-person firm does the work of 1,000.
included for completeness · they doubt itDevelopment automates; humans still steer
100-person companies doing the work of tens of thousands — revolutionary, but turnable to harm (population-scale surveillance, tailored manipulation). Bound by Amdahl’s law: speeding one part shifts the bottleneck — which is exactly why human code review became Anthropic’s new chokepoint.
★ they think we’re likely heading hereAI designs and refines its own successors
Progress paced only by compute. Humans move to oversight of an expanding “virtual lab.” The future they understand least — especially whether alignment holds, or whether rare misalignments compound as models build successors, until control slips.
the one they’re most uncertain aboutBuild the option to slow down — verifiably
The piece ends on policy, not product. A unilateral pause just changes who leads; what’s missing is the ability to verify others have actually slowed.
Why a credible pause is hard — and worth building toward
A slowdown that only lets the least cautious catch up leaves everyone less safe. So the goal is the option: systems that let frontier labs verify others have genuinely stopped. Anthropic says if such systems existed and peers paused verifiably, it expects it would too.
Detection beats verification — and even that’s tough
Training runs are easier to conceal than missile silos, inputs are general-purpose, and whoever continues while others pause inherits the lead.
We’ve done it before — slowly
Regimes like the INF Treaty built verification and trust over decades. The authors’ blunt line: “We don’t have that long.”
Reading it in proportion
- This is one lab’s account of its own internal data — much previously unreported, not independently audited.
- The soft spots are stated in the original: lines-of-code overstates productivity; the self-reported 4× is probably high; the headline research result didn’t transfer to production scale; the next-step test used cherry-picked moments.
- “More autonomous” is not “fully autonomous” — every standout result still had a human framing the problem and defining success.
- That the authors surface these caveats themselves — against their own incentive — is part of what makes the document serious.
Why It Matters
The report matters because it frames recursive self-improvement as a practical engineering question rather than a distant theory. If AI systems can not only perform assigned work but also choose useful research directions, the pace of AI development could become tied more closely to available compute than to human labor cycles.
For policymakers, companies and safety researchers, that would change the planning problem. Oversight systems built around human review, slow release cycles and predictable research timelines may be less effective if models can compress experiments, code changes and iteration into much shorter periods.
Background
Recursive self-improvement refers to a loop in which an AI system helps improve the next version of itself or the systems that build it. The Anthropic Institute piece argues that AI is already taking over parts of the “doing” of AI research: writing code, running experiments and producing candidate results.
The unresolved question is whether AI can take over more of the “deciding”: choosing which problems matter, which experiments are worth running and which findings can be trusted. The report compares this to the difference between a junior employee completing assigned work, an experienced engineer designing an approach and a senior researcher deciding what the team should work on next.
“AI is already, measurably, accelerating the development of AI.”
— Anthropic Institute, as summarized by Thorsten Meyer AI
“Not here yet, not inevitable.”
— Thorsten Meyer AI summary of the Anthropic Institute piece
What Remains Unclear
It is not yet clear whether the reported gains will transfer cleanly to production-scale frontier-model research. The weak-to-strong supervision result cited by Anthropic did not transfer cleanly to larger production models, according to the source material.
It also remains unproven that AI systems can replace human research taste: deciding which problems to pursue, when results are reliable and when a project should be abandoned. That is the main bottleneck the report identifies.
What’s Next
The next milestone is whether AI agents can move from executing well-scoped research tasks to reliably setting research direction. Future evidence will likely focus on whether agents can pick productive experiments, validate their own findings and improve systems under conditions closer to real frontier-model development.
Key Questions
Does the report say recursive self-improvement is already happening?
No. The report argues that AI is already helping build AI, but it says full recursive self-improvement has not been reached.
What evidence does Anthropic cite?
The evidence includes public benchmark trends from METR, SWE-bench and CORE-Bench, plus internal Anthropic figures on Claude-assisted coding and an April 2026 agent research test.
What is still handled by humans?
According to the report, humans still set goals, choose research problems, judge results and define evaluation frames in key cases.
Why does this matter now?
If AI can automate more of AI research, development timelines could shorten and existing oversight processes may need to adapt faster than expected.
Source: Thorsten Meyer AI
