📊 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 presents new data showing AI models are increasingly capable of automating AI development tasks. While human decision-making remains a bottleneck, the evidence suggests self-improving AI could emerge sooner than expected.
Anthropic has released new internal data indicating that AI models are significantly accelerating their own development processes, raising the possibility of recursive self-improvement if current bottlenecks are overcome. This development is based on measurable progress in public benchmarks and internal metrics, suggesting AI is already increasingly capable of automating key aspects of research and coding.
The report from The Anthropic Institute details how AI models like Claude have shown rapid improvements in performing tasks traditionally done by humans, such as coding and experimentation. For example, the proportion of code written by Claude increased from single digits to over 80% within fifteen months, and benchmarks tracking AI’s ability to handle complex research tasks show a doubling of capability roughly every four months.
These metrics are reinforced by public data, such as METR, which tracks AI’s ability to complete increasingly complex software tasks. Tasks that once took days for humans are now within reach of AI models in hours or less, and this trend is expected to continue. However, internal data reveal that while AI excels at execution, decision-making—such as choosing which problems to pursue—remains a significant gap.
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 coding assistant 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.
machine learning development platform
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 Systems Performance Engineering: Optimizing Model Training and Inference Workloads with GPUs, CUDA, and PyTorch
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.
Potential for AI to Self-Accelerate Development
This evidence suggests that AI systems are already capable of automating substantial parts of their own development. If the current bottleneck—human decision-making—can be automated or minimized, AI could enter a loop of recursive self-improvement. Such a scenario could dramatically accelerate technological progress, raising important questions about control, safety, and the future pace of AI advancement.
From Benchmarks to Internal Data on AI Progress
Historically, claims about AI self-improvement have been speculative, but recent data from Anthropic provides concrete evidence of rapid progress. Public benchmarks like METR, SWE-bench, and CORE-Bench show consistent patterns of acceleration, with capabilities doubling every few months. Internally, Anthropic’s data indicates that AI models have become dominant contributors to coding and research tasks within the lab, marking a shift toward greater automation of AI development itself.
This trend raises the possibility that, with further advances, AI could autonomously design its successors, a key step toward recursive self-improvement. Nonetheless, experts emphasize that the decision-making layer—determining which problems to solve—is still human-controlled, and whether this bottleneck can be overcome remains uncertain.
“The data from Anthropic shows that AI is already automating a significant portion of its development, but the critical bottleneck is decision-making, which is still largely human-driven.”
— Thorsten Meyer, AI researcher
Unresolved Questions About Autonomous AI Development
While the data shows rapid progress in automation of coding and experimentation, it remains unclear whether AI can autonomously decide which problems to pursue or design entirely new architectures without human input. The key bottleneck—human decision-making—has not yet been demonstrated to be automatable at scale. Experts warn that the transition from improved execution to full self-directed development is not guaranteed and may face unforeseen technical or safety challenges.
Next Steps for Monitoring AI Self-Improvement Progress
Researchers and industry observers will closely monitor internal developments at labs like Anthropic, looking for signs that decision-making capabilities are also becoming automated. Public benchmarks may continue to evolve, providing more data on AI’s capabilities. Additionally, discussions around safety, control, and regulation are expected to intensify as the possibility of recursive self-improvement becomes more tangible.
Key Questions
Can AI currently fully automate its own development?
No, AI models are improving in automating coding and experimentation tasks, but human decision-making in choosing research directions remains a bottleneck.
What is recursive self-improvement?
It is a process where AI systems autonomously improve their own capabilities, potentially leading to rapid, exponential progress.
Why is the decision-making bottleneck important?
Because automating the choice of research goals and design directions is critical for AI to fully self-improve without human intervention.
What are the risks if AI begins self-improving rapidly?
Potential risks include loss of control, unintended behaviors, and safety concerns, which are topics of ongoing debate and research.
When might autonomous AI self-improvement become a reality?
It is uncertain; current data suggests it could happen if the decision-making bottleneck is overcome, but no specific timeline can be confidently predicted.
Source: ThorstenMeyerAI.com
