Breaking

Anthropic's Clark Puts 60% Odds on Recursive Self-Improvement by End of 2028

June 22, 2026 at 19:00 EDT

Foundation Models
Policy & Regulation
AI Agents

Anthropic co-founder Jack Clark and the company's head of economics, Peter McCrory, appeared on Bloomberg's Odd Lots podcast to discuss AI's impact on the labor market and society, existential risk, and preparations for recursive self-improvement. The roughly 66-minute episode was released on June 19, 2026, with hosts Joe Weisenthal and Tracy Alloway leading the conversation.

June 19, 2026 · Anthropic — Frontier Research

Claude Now Writes Over 80% of Anthropic's Own Code

Co-founder Jack Clark and Head of Economics Peter McCrory describe a frontier lab where AI already accelerates its own development — and quietly prepares for the next step: recursive self-improvement.

80%+

of code merged into Anthropic's own codebase is written by Claude (as of May 2026)

~8×

engineer code output vs. 2021–2025 after adopting the Claude Code agent

RSI

recursive self-improvement: not autonomous yet — preparatory stage

Engineer code output: before vs. after Claude Code

Each block ≈ 1× of baseline output

1×

2021–2025

~8×

with Claude Code

An eightfold jump in output — read as early evidence that AI is already speeding up its own development.

The path toward self-improvement

AI writes code

80%+ of the codebase

→

Faster R&D

~8× engineer output

→

Designs successors

approaching, with compute

Fully autonomous RSI is not yet reached — but with sufficient compute, the firm says systems are nearing a stage where they could design and develop the next generation.

Developer view

A dramatic productivity gain. Claude Code is becoming a staple of everyday development workflows.

Open questions

A development "pause" looks unrealistic under competition — global coordination is seen as essential. Hallucinations and complex tasks remain limits.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.