📊 Full opportunity report: When AI Builds Itself: Inside Anthropic’s Evidence on Recursive Self-Improvement on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Anthropic’s latest report provides concrete data indicating AI systems are increasingly capable of automating parts of their own development. While not yet autonomous in designing their successors, the evidence suggests rapid progress that could lead to recursive self-improvement if certain human-controlled steps are automated.
Anthropic has released new data indicating that AI systems are increasingly capable of automating significant parts of their own development, a development that could lead to recursive self-improvement if certain human oversight is automated. The findings, based on internal metrics and public benchmarks, suggest that AI’s ability to write code and conduct experiments is advancing rapidly, though the crucial decision-making step remains human-controlled for now.
The report from Anthropic’s Institute presents concrete, internally sourced data showing that AI models are now capable of automating a growing share of their own development tasks. For example, as of May 2026, over 80% of code integrated into Anthropic’s projects was authored by their AI model Claude, up from just a few percent in early 2025. Public benchmarks like METR, SWE-bench, and CORE-Bench show a steady acceleration in AI’s ability to perform increasingly complex tasks, such as fixing bugs, reproducing research results, and handling longer, more demanding tasks. The core insight is that AI can already automate much of the ‘doing’ of AI research—writing code, running experiments, producing results—but the ‘deciding’—choosing which problems to tackle—is still primarily human. The authors emphasize that while this progress is measurable and significant, full recursive self-improvement, where AI autonomously designs and improves its own successor systems, remains a future possibility, not an imminent reality.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.
Vibe Coding: Building Production-Grade Software With GenAI, Chat, Agents, and Beyond
As an affiliate, we earn on qualifying purchases.
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.
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 software
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
AI research automation software
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.
Implications of Accelerating AI Self-Development
This evidence suggests that AI systems are rapidly closing the gap in automating their own development processes, which could accelerate progress in AI capabilities significantly. If the ‘taste’ or strategic decision-making step—currently human-controlled—becomes automated, AI could enter a loop of self-improvement running at the speed of compute. This raises questions about future control, safety, and the pace of technological change, making it a crucial development for researchers, policymakers, and industry leaders to monitor.
Progress in AI Capabilities and Benchmark Trends
Anthropic’s report builds on observable trends in public benchmarks like METR, SWE-bench, and CORE-Bench, which show AI models rapidly improving in tasks such as coding, bug fixing, and reproducing research results. These benchmarks have demonstrated near-exponential growth in AI proficiency over the past two years, with models now capable of handling tasks that previously required days of human effort. Internally, Anthropic has tracked a dramatic increase in the volume of code and experiments driven by AI, indicating an acceleration in AI-driven research and development. The report underscores that these developments are happening now, based on measurable data, rather than future speculation.
“The data shows AI is already automating significant portions of its own development, and the pace of this progress is accelerating.”
— Thorsten Meyer, author of the report
Uncertainties Surrounding Autonomous AI Self-Improvement
It remains unclear whether AI will soon be able to fully automate the goal-setting and strategic decision-making aspects of research, which are currently human-led. The report emphasizes that while AI can automate coding and experimentation, the step of choosing which problems to pursue is still controlled by humans. Whether this gap will close in the near future, enabling true recursive self-improvement, is an open question. Additionally, the implications for safety, control, and unintended consequences are still being debated and are not yet fully understood.
Monitoring AI Development and Preparing for Potential Self-Improvement
The next steps involve continued measurement of AI capabilities, internal testing of autonomous decision-making, and discussions around safety protocols. Researchers and industry leaders will need to assess whether AI systems can or should be given greater autonomy in research tasks. Policymakers may also consider regulations to address the possibility of rapid, self-driven AI progress. The ongoing collection of data and transparency from labs will be crucial in understanding whether the current trends lead toward autonomous self-improvement or remain within human oversight.
Key Questions
What is recursive self-improvement in AI?
Recursive self-improvement refers to AI systems that can autonomously improve or redesign themselves without human intervention, potentially leading to rapid, exponential progress in capabilities.
How close are we to AI automating its own development?
According to Anthropic’s data, AI is already automating significant parts of its development, such as coding and experimentation, but the strategic decision-making step remains human-controlled. Full automation of self-improvement is still a future possibility.
What are the risks of AI self-improvement?
Potential risks include loss of human oversight, unpredictable behavior, and rapid technological change that could outpace safety measures. These concerns are actively discussed among researchers and policymakers.
Will AI self-improvement happen soon?
It is uncertain. While current trends suggest rapid progress, the critical step of autonomous goal-setting and strategic decision-making has not yet been achieved. Experts warn it could happen sooner than expected if certain bottlenecks are removed.
What should be done to prepare for AI self-improvement?
Ongoing research, transparency, safety protocols, and policy development are essential to prepare for potential future scenarios involving autonomous AI self-improvement.
Source: ThorstenMeyerAI.com
