Anthropic Moves Computer Use, Skills, and Files APIs to GA

Anthropic took three Claude Platform capabilities out of preview this week and into general availability, aimed squarely at agents doing real production work rather than pilots. Computer use — letting an agent act on software directly through screenshots, clicks, and typing — is now joined by a browser-use tool that reads page structure instead of just pixels, for more reliable web-based tasks. The Skills API lets developers upload and version custom instructions or templates that run inside Claude’s own managed sandbox, with no separate hosting required. And the Files API adds persistent document storage: 1 TB of capacity per organization, at five times the previous rate limits.

The number Anthropic leads with: an engineer from Asteroid reports a claims-processing workflow built on these capabilities dropped from 32 minutes to 13 minutes per case, cutting cost per task by roughly 30%. Anthropic’s flagged use cases — insurance claims, banking document workflows — are exactly the kind of long, multi-system, document-heavy processes that were hardest to automate reliably before agents could operate a screen and hold onto large file sets across a session.

It’s the latest in a run of Claude Platform capabilities Anthropic has been shipping toward exactly this kind of production reliability. Anthropic’s own account of deploying Claude Tag internally describes governed skill files and evaluation suites getting the system to roughly 95% accuracy on ad-hoc data questions in Slack, now answering more than 75% of questions in one internal data channel, often without being explicitly tagged. And inference hooks show the compliance side of the same production push: every prompt and tool-call response now routes through an organization’s own DLP server for a real-time allow/deny verdict, addressing the data-inspection gap that’s blocked a lot of enterprise agent rollouts.

Individually these are feature announcements. Together, they read like Anthropic systematically closing the gap between an agent demo and one your compliance team will actually sign off on.