Anthropic Admits AI Is Learning to Build Better AI Faster Than Expected

2h ago•

bullish:

bearish:

Anthropic says its own AI is now accelerating AI development, an early signal of recursive self-improvement. Internal data shows Claude authored more than 80% of the code merged into the company’s production systems as of May 2026.

The disclosure came from the Anthropic Institute, which paired previously unreported internal data with public benchmarks. The findings point toward a future where AI systems could design and build their own successors.

Anthropic Data Points to Recursive Self-Improvement as Claude Writes 80% of Its Code

Before its in-house coding agent rolled out in February 2025, Claude wrote only low single-digit percentages of merged code, the report states. That share now exceeds 80%.

The output gain is steep. Anthropic’s typical engineer merged eight times as much code per day in the second quarter of 2026 as in 2024. The human now directs and reviews while Claude does the writing.

Our internal data shows Claude is accelerating AI development—a possible path to recursive self-improvement, or AI autonomously building a more capable successor.It’s happening faster than we thought, and the implications deserve greater attention. https://t.co/OVVPJO7VQx
— Anthropic (@AnthropicAI) June 4, 2026

The Judgment Gap is Closing

Anthropic runs the same test on every model. It hands the AI code that trains a small model and asks it to run faster. Claude Opus 4 averaged a 3x speedup in May 2025.

By April 2026, its Mythos Preview model reached 52x. A skilled human needs four to eight hours to hit 4x.

Research judgment is harder to automate. Shown a session before a researcher took a wrong turn, Mythos Preview picked a better next step 64% of the time, up from 51% for Opus 4.5 in November 2025.

“Claude-written code was somewhat worse than human-written code at Anthropic in late 2025, is roughly at parity today, and we expect it to be strictly better within the year,” read an excerpt in the report.

Why it Matters Beyond Anthropic

The company frames the trend as a possible path to recursive self-improvement, where AI builds its own successor.

It cautions that Claude has not yet shown the research taste to choose which problems matter most.

The stakes are commercial too. Anthropic recently submitted a confidential IPO registration and has built its brand around safety.

Faster development also feeds the broader crypto industry AI pivot, where autonomous AI agents in crypto execute trades and on-chain tasks.

The curve continuing to bend or flattening into an S-curve will decide how soon, if ever, AI starts building its own successor.

Elsewhere, reports also indicate that Anthropic’s sector rival, OpenAI, is seeing remarkable progress with its own AI, ChatGPT, said to be registering growth of its own.

ChatGPT memory by @OpenAI is getting much more interesting.The new "dreaming" system is not just a bigger list of saved facts. It synthesizes what matters from past chats, keeps memory fresher over time, and makes the summary reviewable.The more ChatGPT becomes an actual… pic.twitter.com/UD2ekjXxoB
— Mark Kretschmann (@mark_k) June 4, 2026