On August 20, 2026, Anthropic announced that Computer Use, the Skills API, and the Files API are generally available on the Claude Platform, with a new browser use tool added to Computer Use. The significance is the combination: an agent can operate software that has no API, load an organization's instructions, scripts, and templates when needed, and read or produce durable file artifacts.
Computer Use lets an agent click, type, and scroll from a screenshot, allowing it to work with desktop or web software that was not designed for automation. The new browser use tool still uses the visual page, but also reads page structure so the agent can target a field or button instead of relying only on pixel positions. For legacy back offices, vendor portals, and internal tools, this creates an automation path that does not require an immediate API rewrite.
The Skills API addresses how team expertise enters an agent. Anthropic describes a skill as a folder containing instructions, scripts, and templates that Claude loads only when a task calls for it. Teams can upload and version their skills, attach them to requests, and run them in Claude's code execution sandbox without hosting a separate skill service.
The Files API provides a reusable document layer. A team can upload a PDF or spreadsheet once, reference it by file ID in later requests instead of sending it again, and download files created by the agent. Anthropic's example is a claims workflow: the agent reads an intake document from the Files API, follows a filing procedure encoded in a skill, completes a form in an insurer portal through browser use, and saves the confirmation as a file.
Anthropic also lists multi-action turns for Computer Use, page structure in browser use, versioning for the Skills API, and automatic expiration, higher rate limits, and organization storage for the Files API. The company says the Skills API and Files API are available through Microsoft Foundry, while updated computer use and browser use tools are coming to Google Cloud Vertex AI. These are availability statements from the product announcement, not a guarantee that every account or region has the same access at the same time.
For AI workflow design, the release separates the agent loop into three layers: Computer Use takes action in a real interface, Skills turn organizational method into versioned modules, and Files retain inputs and outputs. A production team should define browser permissions, sensitive-data boundaries, skill ownership, file lifecycles, and audit records for each action rather than stopping at a successful demo.
General availability does not mean an agent can be left unattended in production. Browser automation can encounter changed layouts, insufficient permissions, duplicate submissions, and wrong context; scripts inside skills still need code review, rollback, and test data. Anthropic's page includes customer descriptions of speed, cost, and completion outcomes, but those are company-published examples rather than independent cross-workflow benchmarks.
The clearest signal is that enterprise agents are moving from answering questions to completing work across software, expertise, and files. Safe deployment will depend on whether an organization puts interface actions, professional rules, document retention, and human approval points into one traceable workflow.
