Anthropic has moved several Claude Platform agent capabilities from earlier availability into general availability, making computer use, the Skills API and the Files API production-facing parts of its developer platform as of August 20, 2026. The company also added a browser-use tool for agents that work inside web applications.
The update is not just another model release. It changes the shape of what developers can ask an AI platform to do through an API. Instead of sending prompts to a text model and separately wiring every file, workflow and browser action in application code, teams can now use Claude with more persistent files, versioned skills and higher-level computer and browser controls.
For companies building agent products, internal automation, document workflows or web-portal assistants, that makes the Claude Platform look more like an agent runtime. It also creates new governance work for any AI API gateway, because routing decisions increasingly depend on tool permissions, file retention, browser access and skill versions — not only model name, price and latency.
What changed in the Claude Platform release
Anthropic says computer use, the Skills API and the Files API are now generally available on the Claude Platform. The updated computer use tool can take several actions per turn, rather than requiring one model call for each action. That matters for latency and reliability: long workflows that previously required many round trips between model and application can be compressed into fewer exchanges.
The new browser-use tool is aimed at agents operating in web applications. Anthropic’s distinction is meaningful. Traditional computer-use systems often reason over pixels and desktop-like actions. A browser-focused tool can give agents a more structured way to work with web pages, reducing some of the brittleness that comes from treating every interface as a screenshot.
The Skills API gives developers a way to upload and version skills, then attach those skills to requests. In practice, a skill can package instructions, domain-specific procedures or reusable capabilities that a team wants an agent to apply consistently. Versioning is the key production feature: without it, teams have a hard time knowing whether two runs used the same operating procedure.
The Files API addresses another common bottleneck in agent applications. Developers can upload files once, reference them later by ID and download generated files. Anthropic says the updated Files API includes automatic expiration, five-times higher rate limits and 1 TB of storage per organization. For document-heavy workflows — legal review, financial analysis, support operations, healthcare administration or back-office automation — this reduces the need to repeatedly upload the same material with every request.
Why this matters for developers and businesses
The release pushes agent development further away from stateless chat completions. A production agent now needs memory-like file persistence, reusable skills, permissioned tools and auditability around every action it takes. That is a different operating model from simply swapping one model endpoint for another.
Developers are likely to feel the effect first in workflow design. A document-processing agent can keep a source file available by ID, apply a specific version of a review skill and then return generated files without custom upload plumbing for every step. A web-operations agent can use browser tooling to work inside SaaS applications. A desktop-style automation agent can complete multiple actions in one turn, potentially reducing orchestration overhead.
Businesses will feel it in policy and risk management. Computer use and browser use are powerful, but they also broaden the blast radius of a bad instruction, compromised account or poorly scoped integration. Teams will need to decide which users, projects and environments can enable these tools, what files may be persisted, how long data should remain available and how skill versions are approved.
This is where gateways and internal AI platforms become more important. A multi-model API layer can no longer treat all Claude requests as equivalent. One request might be a normal text generation call. Another might involve a persistent file, a privileged browser session and a versioned automation skill. Those are materially different from a security, cost and compliance perspective.
Implications for API gateways and agent infrastructure
For products such as Model Gate, the practical connection is clear: agent capabilities are becoming routing and governance dimensions. A gateway that already handles unified billing, API-key management, usage analytics and team controls may need to expose additional policy controls for files, skills and tool use.
At minimum, teams should expect demand for per-tool permissions: who can use computer use, who can invoke browser use, which API keys can attach skills and which projects can store files. Usage analytics also become more nuanced. Token counts and model costs remain important, but operators may also want to know which workflows used persisted files, which skill versions were invoked and whether browser-enabled agents are driving unusual spend or risk.
Partner and reseller platforms face a similar issue. If a service is built on top of Claude through a partner API or gateway, the end customer may not see the underlying Anthropic console. That means the reseller platform may need to provide its own controls for file lifecycle, skill updates and tool approvals. Otherwise, the most consequential parts of the agent workflow are hidden from the business that is accountable for them.
The release also fits a broader industry pattern. Recent developer-platform updates have focused less on raw model access and more on agent packaging, tool protocols, code-review automation, source connectors and cost observability. Anthropic’s update adds another strong signal: the competitive surface is moving from “which model answers best” toward “which platform can safely execute work.”
What remains uncertain
The strongest verification for this release comes from Anthropic’s own announcement and documentation surfaces. There is limited independent reporting so far, and real-world performance will depend on how developers implement permissions, error handling and evaluation around the tools.
Availability across clouds is also uneven. Anthropic says the Skills API and Files API are available through Microsoft Foundry. It also says updated computer use and browser-use tools are coming soon to Google Cloud Vertex AI. Teams standardizing on a specific cloud will need to check exact regional availability, platform controls and enterprise policy support before planning migrations.
Regulated organizations should be especially careful. File persistence, generated-file downloads and agents that can operate web or computer interfaces may interact with data-handling obligations, audit requirements and sector-specific rules. Anthropic’s general availability label is a production signal, not a substitute for a customer’s own compliance review.
The direction, however, is clear. Agent APIs are becoming less like isolated inference endpoints and more like managed execution environments. That gives developers more leverage, but it also means businesses need stronger controls over what agents can see, store and do.