📊 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 recent report provides data indicating AI systems are already automating significant parts of their own development. While full recursive self-improvement is not yet happening, the evidence suggests it could occur sooner than expected if certain bottlenecks are removed.
Anthropic has released a detailed analysis indicating that AI systems are already automating substantial portions of their own development, with evidence suggesting the potential for recursive self-improvement if current bottlenecks are eliminated. This development could accelerate AI progress significantly, raising important questions about future capabilities and safety.
The report from The Anthropic Institute emphasizes that AI models, particularly those from Anthropic, are increasingly capable of performing tasks traditionally done by humans in AI research and engineering. Data shows that AI systems like Claude now generate over 80% of code merged into the company’s projects, a dramatic increase from early 2025.
Public benchmarks such as METR, SWE-bench, and CORE-Bench reveal that AI models are rapidly advancing in ability, doubling their task completion horizon roughly every four months. For example, tasks that once took skilled humans days could be within AI’s reach as early as this year, with longer-term tasks approaching feasibility in the next few years.
Inside labs, the data suggests that AI is already capable of autonomously designing methods for engineering problems and executing experiments, though it still struggles with higher-level decision-making—namely, choosing which problems to pursue or which results to trust. The authors highlight that this gap is the critical barrier before true recursive self-improvement could occur.
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.
AI code generation tools
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
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.
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.
autonomous AI development platforms
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 experiment automation tools
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 AI Automating Its Own Research and Development
This evidence suggests that AI systems are already changing the pace of development within labs, potentially reducing the time and human effort needed to advance AI capabilities. If the bottleneck of human decision-making and goal-setting can also be automated, it could lead to a rapid, self-reinforcing cycle of improvement—raising both opportunities and risks for the future of AI.
Understanding whether this process is imminent or still distant is crucial for policymakers, researchers, and companies preparing for the next phase of AI evolution. The possibility of AI systems designing their own successors could accelerate technological progress but also complicate safety and control measures.
Current State of AI Self-Development Capabilities
The notion of recursive self-improvement has long been discussed in AI safety and development circles, but until now, it has largely been speculative. Recent data from Anthropic, however, indicates that AI models are already performing many tasks involved in AI research—such as coding, testing, and debugging—at levels that suggest a significant acceleration in internal development processes.
Public benchmarks have documented rapid progress, with models moving from low success rates to near-saturation in tasks like reproducing research results and fixing bugs within a short period. These trends point toward a future where AI could take on more decision-making roles in research, provided the remaining gaps are closed.
“The data from Anthropic indicates that AI systems are already automating a large part of their own development, which could lead to a self-improving loop if the bottleneck of human decision-making is removed.”
— Thorsten Meyer, AI researcher
Uncertainties Surrounding AI Self-Improvement Timing
It remains unclear whether current trends will continue at the same pace, or if technical, safety, or ethical challenges will slow progress. The authors acknowledge that full recursive self-improvement is not yet happening and that the transition depends on overcoming significant decision-making bottlenecks in AI systems.
Additionally, there is uncertainty about how quickly AI systems could autonomously design their successors once these bottlenecks are addressed, and what safety measures would be necessary to manage such capabilities.
Next Steps in Monitoring AI Self-Development Progress
Researchers and industry stakeholders will likely focus on further measuring AI’s capabilities in autonomous research tasks and developing safety frameworks for increasingly autonomous AI systems. Key milestones include demonstrating AI’s ability to fully automate goal setting and decision-making in research contexts.
Monitoring internal development metrics and benchmarks will be essential, alongside policy discussions about managing the risks associated with potential self-improving AI systems.
Key Questions
Could AI systems soon design their own successors?
Current evidence suggests that AI is approaching the capability to automate parts of its own research and development, but fully autonomous design of successors remains a future possibility dependent on overcoming existing decision-making bottlenecks.
What are the main barriers to recursive self-improvement?
The primary challenge is automating the high-level decision-making process—choosing which problems to pursue and which results to trust—currently a task that still requires human judgment.
How reliable are the benchmarks used to measure AI progress?
Benchmarks like METR, SWE-bench, and CORE-Bench provide valuable data on AI capabilities, but they do not directly measure internal development speed or the potential for recursive self-improvement. Internal data from labs offers deeper insights but is less publicly available.
What are the safety concerns with AI self-improvement?
If AI systems begin to autonomously improve themselves, it could lead to rapid, unpredictable advancements that challenge existing safety and control mechanisms. Careful oversight and new safety protocols will be necessary.
Source: ThorstenMeyerAI.com
