Autoresearch Is Becoming a Software Engineering Loop

Four months after Karpathy's repo, autoresearch reads as a software-engineering loop: a goal, a hard evaluator, and an agent that iterates overnight.

July 9, 2026

Fable 5's Better Signal Is Boring Work That Ships

Claude Fable 5's demos dazzle, but its stronger early signal is unglamorous, reviewable work that ships: release review, research tooling, and a large migration.

July 6, 2026

Anthropic's Own Data Says the Coding-Agent Bottleneck Is Now the Human

Anthropic's study of 400,000 Claude Code sessions: humans make ~70% of planning decisions, Claude ~80% of execution — and success tracks domain expertise.

July 5, 2026

Microsoft Study Finds Coding-Agent Adopters Merged 24% More Pull Requests

A study of Microsoft's own rollout of Claude Code and Copilot CLI estimates a durable output lift — and warns that merged PRs are a proxy, not delivered value.

July 5, 2026

Alibaba Bars Staff From Using Claude Code Over Anthropic Tracking Code

After Anthropic used code to flag China-linked users, Alibaba called Claude Code high-risk and pointed staff to its own Qoder.

July 4, 2026