Issue 1 — memory everywhere, agents in the browser

The sweep’s first week in one read: the assistant’s state followed you everywhere, its hands reached into the browser, and the API layer under all of it went generally available — three moves that were each preview or rumor a month ago.

The picks

• Claude’s memory now works everywhere — memory spans chat, Cowork, and the cloud, with every stored topic visible and editable in Settings. This is the week’s real shift: the assistant stops being surface-local. The support release notes add the sensitive-topics control, and TestingCatalog saw the Cowork rollout reach Free plans.

• Cowork agents get a built-in browser — agents can now open and act on web pages directly instead of working only through files and connectors. Paired with Claude in Chrome going generally available, the read-the-web layer is no longer the bottleneck it was all year.

• The API platform went stable in one wave — computer use dropped its beta header, a browser-use tool launched for driving a hosted browser by reading the page’s accessibility tree, and the official production-agents guide walks one build through computer use, the Skills API, and the Files API — all three now stable. The Admin API reached the ant CLI and every SDK the same week.

• A million agent-written lines, running in production — Paul Dix, quoted by Simon Willison: AI wrote a million lines and refined them over months into software now on millions of developer machines. The frank version of the number everyone keeps guessing at.