Anthropic Admits AI Is Learning to Build Better AI Faster Than Expected
Anthropic says its personal AI is now accelerating AI growth, an early sign of recursive self-improvement. Internal information reveals Claude authored greater than 80% of the code merged into the corporate’s manufacturing methods as of May 2026.
The disclosure got here from the Anthropic Institute, which paired beforehand unreported inside information with public benchmarks. The findings level towards a future the place AI methods may design and construct their very own successors.
AI Now Writes Most of Anthropic’s Code
Before its in-house coding agent rolled out in February 2025, Claude wrote solely low single-digit percentages of merged code, the report states. That share now exceeds 80%.
The output acquire is steep. Anthropic’s typical engineer merged eight occasions as a lot code per day within the second quarter of 2026 as in 2024. The human now directs and opinions whereas Claude does the writing.
Follow us on X to get the most recent information because it occurs
The Judgment Gap is Closing
Anthropic runs the identical check on each mannequin. It fingers the AI code that trains a small mannequin and asks it to run quicker. Claude Opus 4 averaged a 3x speedup in May 2025.
By April 2026, its Mythos Preview model reached 52x. A talented human wants 4 to eight hours to hit 4x.
Research judgment is more durable to automate. Shown a session earlier than a researcher took a incorrect flip, Mythos Preview picked a greater subsequent step 64% of the time, up from 51% for Opus 4.5 in November 2025.
“Claude-written code was considerably worse than human-written code at Anthropic in late 2025, is roughly at parity at present, and we count on it to be strictly higher throughout the 12 months,” read an excerpt within the report.
Why it Matters Beyond Anthropic
The firm frames the pattern as a doable path to recursive self-improvement, the place AI builds its personal successor.
It cautions that Claude has not but proven the analysis style to select which issues matter most.
The stakes are industrial too. Anthropic lately submitted a confidential IPO registration and has constructed its model round security.
Faster growth additionally feeds the broader crypto industry AI pivot, the place autonomous AI agents in crypto execute trades and on-chain duties.
The curve persevering with to bend or flattening into an S-curve will determine how quickly, if ever, AI begins constructing its personal successor.
Subscribe to our YouTube channel to watch leaders and journalists present skilled insights
The submit Anthropic Admits AI Is Learning to Build Better AI Faster Than Expected appeared first on BeInCrypto.
