📊 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 claims that current AI systems are already accelerating AI development by automating coding and experiments. Their internal data suggests that if a key human decision-making step is automated, recursive self-improvement could occur sooner than expected. The evidence is based on public benchmarks and internal metrics, but some uncertainties remain about the timeline and feasibility.
Anthropic has released new internal data suggesting that AI systems are already significantly automating parts of their own development, with the potential to enter a loop of recursive self-improvement if a remaining human-controlled step is automated. This development could accelerate AI progress far faster than traditional human-led research, though experts emphasize it is not yet inevitable.
The Anthropic Institute’s report presents measurable evidence that AI models like Claude are rapidly improving in tasks related to coding and research, with more than 80% of code merged into their systems now authored by AI. Public benchmarks such as METR and SWE-bench show a doubling of AI capabilities every four months, indicating a clear acceleration in AI’s ability to perform tasks traditionally done by humans.
Inside the labs, Anthropic researchers distinguish between engineering work—writing code and infrastructure—and research work—deciding experiments and interpreting results. Their data shows AI already matches or surpasses skilled humans in executing specified experiments but still lags in autonomous goal-setting and strategic decision-making. The authors argue that if AI can automate the ‘taste’ or human judgment in research, a self-improving cycle could begin, running at the speed of compute rather than human effort.
As of May 2026, over 80% of code in Anthropic’s development pipeline is AI-generated, a dramatic increase from early 2025. The trend in benchmark tasks suggests that AI could soon handle tasks that currently take days for humans, with some models capable of 12-hour tasks and projections indicating weeks-long tasks could be within reach by 2027.
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.
Agentic Coding with Claude Code: The everyday developer's guide to agentic coding with Claude Code
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
Patriola's Guide to Claude: Research Navigator: Use Claude to Map a Field and Produce Reproducible Findings
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.
AI-assisted Software Development: A Pragmatic Operating Model for Safe Adoption in Regulated Environments
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
The AI-Driven Leader: Harnessing AI to Make Faster, Smarter Decisions
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 indicates that AI is already making significant strides in automating parts of its own development process, which could lead to a rapid, self-sustaining cycle of improvement. If AI systems can autonomously design and refine their successors, this could drastically shorten the timeline for AI capabilities to surpass human oversight, raising questions about control, safety, and regulation.
While the report emphasizes that full recursive self-improvement is not yet occurring, the possibility that it could happen sooner than most institutions expect makes this a critical development for AI researchers, policymakers, and industry leaders to monitor closely.
Data-Driven Evidence of AI Progress in Self-Development
Anthropic’s report builds on public benchmarks like METR, SWE-bench, and CORE-Bench, which measure AI capabilities in coding, experiment reproduction, and research tasks. These benchmarks show a consistent pattern of exponential growth in AI performance over the past two years, with capabilities doubling roughly every four months.
Internally, Anthropic tracks metrics related to code authorship, experiment execution, and research decision-making. Their data reveals that AI models are increasingly handling tasks that previously required human expertise, with a notable jump in code contribution from AI systems in early 2025. This internal data forms the backbone of their argument that AI is already automating key aspects of its own development process.
However, the report clarifies that the critical bottleneck remains human judgment—deciding which problems to pursue and interpreting results—areas where AI still lags behind human experts. The potential for recursive self-improvement hinges on automating these decision-making steps, which is still an open challenge.
“Our data shows that AI systems are already automating a significant portion of their own development tasks, and the pace of progress suggests that a self-improving loop could emerge if the human decision layer is automated.”
— Thorsten Meyer, author of the report
Uncertainties About Timeline and Feasibility of Self-Improvement
It is not yet clear when or if AI will fully automate the goal-setting and strategic decision-making processes necessary for recursive self-improvement. The report emphasizes that this remains an open challenge, and current models are not yet capable of autonomous system design at scale.
Additionally, the implications of rapid self-improvement—such as safety, control, and unintended consequences—are not addressed in detail, leaving questions about how to manage such a transition.
Next Steps in Monitoring AI Self-Development Capabilities
Researchers and industry leaders will likely focus on developing AI systems that can autonomously set research goals and design their own improvements. Further internal data collection and benchmarking will be necessary to track progress toward full recursive self-improvement.
Regulators and policymakers may begin to consider guidelines for AI autonomy and safety, especially if the pace of capability growth accelerates. Public disclosure and transparency from labs like Anthropic will be crucial to understanding when and how these developments occur.
In the near term, expect continued updates on internal metrics and external benchmark performance, alongside ongoing debates about the risks and benefits of increasingly autonomous AI systems.
]Key Questions
Is AI currently capable of fully automating its own development?
No, current AI models are automating parts of their development, such as coding and experiment execution, but they have not yet achieved full autonomous self-improvement, especially in strategic decision-making.
What is recursive self-improvement in AI?
Recursive self-improvement refers to an AI system’s ability to autonomously improve its own architecture and capabilities without human intervention, potentially creating a rapid cycle of enhancement.
How soon could AI systems begin self-improving at scale?
The report suggests that if the current trends continue and the bottleneck of human decision-making is automated, self-improvement could occur within a few years, possibly as early as 2027.
What are the risks associated with AI self-improvement?
Potential risks include loss of human oversight, unintended behaviors, and rapid capability escalation that could outpace safety measures. Addressing these requires careful regulation and transparency.
What should researchers and policymakers do next?
Monitoring internal metrics, advancing AI autonomy in goal-setting, and establishing safety protocols are key steps. Transparency from labs like Anthropic will be vital for informed decision-making.
Source: ThorstenMeyerAI.com
