Anthropic’s computer use tool, browser use tool, Skills API, and Files API moved to general availability on August 20, 2026, marking a significant step for enterprise teams that want to move beyond prototype-stage AI agents and into production-grade automation of real software workflows.
The four capabilities address different parts of the agent stack, but they are designed to compose: a single agent can read documents from Files API, apply team-specific instructions from a skill, operate a web application through the browser use tool, and write outputs back as files. Together they give enterprise developers a foundation for building agents that span the full span of a knowledge work task, not just isolated steps.
What Went GA and What Changed
Computer Use: Multi-Action Turns and HIPAA Clearance
The original computer use beta launched in late 2024 and let Claude control a computer by taking screenshots and acting on what it saw. The GA version introduces two meaningful changes.
First, Claude can now take several actions per turn rather than one action per model call. That compresses long workflows into fewer round trips, which reduces latency and cuts cost proportionally. A task that previously required 40 model calls to navigate a multi-step form might now finish in 12.
Second, computer use is now eligible for HIPAA-regulated workloads under Anthropic’s Business Associate Agreement. Healthcare and insurance teams that wanted to pilot computer use agents but could not deploy them in environments handling protected health information now have a path to production.
The clinical impact is already visible in early enterprise deployments. Davide Locatelli, a research engineer working on healthcare and insurance systems, reported that their longest claims workflow dropped from 32 minutes to 13 after adopting the updated tool, a reduction of about 59 percent. Cost per task fell roughly 30 percent across every workflow tested, and task completion reached 100 percent, with no changes to the team’s existing prompts.
Browser Use Tool: Acting on Structure, Not Just Pixels
New at GA is a companion tool called the browser use tool. Where computer use identifies click targets by their position on a screenshot, the browser use tool reads the underlying structure of the page and targets specific elements by type or label. The practical difference is reliability: a button that shifts position when the window resizes will still be found by structure even if its pixel coordinates are wrong.
Both tools share the same multi-action turn model, so switching to the browser use tool does not add latency. For agents that spend most of their time inside web applications, the browser use tool is the better starting point. For agents that operate desktop software or applications where page structure is not available, computer use with screenshot-based targeting remains the right choice.
Skills API: Encoding Team Expertise for Agents
The Skills API addresses one of the harder problems in enterprise AI deployment: how to encode your organization’s specific procedures, formats, and constraints into an agent without re-prompting them on every request.
A skill is a folder of instructions, scripts, and templates that Claude loads when a task calls for it. Teams create skills that encode their own processes: the approval checklist for a procurement agent, the citation format for a legal filing agent, the required schema for a data extraction agent. Once a skill is uploaded via the API, it can be versioned and attached to any request by ID.
Skills run inside Claude’s code execution sandbox, so enterprise teams do not need to operate any additional infrastructure to use them. That removes a deployment friction that blocked some teams from operationalizing their prompting work during the beta period.
Matthew Midson, Managing Director of Banking at Box, described using the Skills API to build Box Agent for banking clients: “A skill captures the firm’s credit methodology and approved memo format; Box Agent applies it to the financial statements and deal documents already in Box and produces a source-grounded credit memo for analyst review. Banks get agents for complex workflows without building each one from scratch.”
This pattern, where team expertise is packaged as a versioned skill rather than duplicated across dozens of prompts, scales well in enterprise settings where procedures evolve over time. Updating a skill propagates the change to every agent request that loads it.
Files API: Storage Purpose-Built for Agents
The Files API provides persistent storage for the documents agents consume and produce. At GA, it adds automatic file expiration (so unused files do not accumulate indefinitely), 5x higher rate limits (relevant for teams running high-volume document workflows), and 1 TB of storage per organization.
The core workflow the Files API enables is simple but high-value: upload a PDF, spreadsheet, or other document once, then reference it by ID in subsequent requests rather than re-transmitting the full file. For an agent processing thousands of documents per day, eliminating redundant uploads materially reduces both latency and ingress cost.
Agents that create outputs (a generated report, a completed form, a processed dataset) can write those files back through the same API, giving human reviewers a consistent place to retrieve and audit agent work.
Why These Four Capabilities Compose
The claims workflow that Anthropic described in the launch announcement is a useful illustration. An insurance claims agent reads the intake document from the Files API, follows a skill that encodes the team’s filing procedure, completes the claim submission in an insurer’s web portal using the browser use tool, and saves the confirmation receipt back as a file.
Each tool handles one boundary that agents previously crossed poorly. Computer use and the browser use tool handle the application boundary: the systems with no API. The Skills API handles the expertise boundary: the know-how that lives in documentation or in people’s heads. The Files API handles the document boundary: the content that moves through the workflow. Code execution and web search, already GA before this launch, handle computation and information retrieval.
The result is a stack where an agent can work in roughly the same loop a human knowledge worker would, moving between documents, systems, and codified procedures, without requiring any of those systems to expose a structured API.
What This Means for Enterprise AI Teams
For teams that have been running computer use in beta, the GA release with HIPAA eligibility and multi-action turns removes the two most common blockers to scaling: compliance clearance and cost at volume.
For teams that have not yet built with computer use, the browser use tool offers a more reliable entry point for web-based workflows, and the Skills API offers a cleaner path to operationalizing team-specific agent behavior than managing complex system prompts.
| Capability | Key GA Change | Relevance for Enterprise |
|---|---|---|
| Computer use | Multi-action turns, HIPAA-eligible under BAA | Scales to high-volume workflows; unlocks healthcare and insurance |
| Browser use tool | New at GA: structure-aware web actions | More reliable than pixel-only targeting in web apps |
| Skills API | Simpler upload and versioning; runs in Claude sandbox | Package team expertise without maintaining extra infrastructure |
| Files API | 5x rate limits, 1 TB per org, auto-expiration | Document-heavy workflows without repeated file uploads |
The platform availability picture is also broadening. Skills API and Files API are available through Microsoft Foundry as well as directly on the Claude Platform, which matters for enterprises that run their AI workloads through Azure under data residency or compliance agreements. Computer use and the browser use tool are coming to Google Cloud’s Vertex AI, extending the options for teams on Google infrastructure.
Practical Starting Points
For teams evaluating where to start, a few patterns are worth testing first.
Document-heavy approval workflows (loan applications, claims, permit filings) are a natural fit for the Files API plus browser use combination. The agent can hold the source document in persistent storage, load the team’s processing skill, and complete the downstream form submission without any custom API integration.
Internal tool automation is a lower-risk entry point for teams newer to computer use. Agents that navigate internal dashboards, pull reports, or enter data into internal systems operate in controlled environments where errors are easier to catch and the HIPAA considerations may not apply.
Skills that encode compliance steps or formatting requirements deliver compounding value. A skill created for one agent can be reused across multiple agents that share the same procedure, and updating it propagates consistently rather than requiring prompt edits across every deployment.
Existing beta integrations keep working while teams migrate to the GA endpoints, so there is no forced cutover deadline.
Connecting the GA Launch to Anthropic’s Broader Trajectory
This launch arrives shortly after Anthropic published its August 2026 risk report and introduced Model 2, and shortly after Claude became available on Microsoft Foundry with expanded agentic capabilities. The pattern is consistent: Anthropic is moving its most capable agentic tools from research-grade beta to production-grade GA, while simultaneously expanding the platforms through which enterprise teams can access them.
The addition of HIPAA eligibility for computer use, in particular, signals that Anthropic views regulated industries as a primary market rather than an edge case. The combination of Skills API’s team-expertise packaging and Files API’s document persistence also addresses a complaint that has been common among enterprise teams: that AI agents are good at one-off tasks but struggle to maintain context and procedure across complex, multi-step workflows.
For teams building computer use agents for enterprise automation, the GA release with multi-action turns and HIPAA clearance changes the production calculus significantly. What previously required careful scoping to avoid compliance exposure now has a supported path for deployment in the most regulated enterprise environments.
The full documentation for computer use, the browser use tool, the Skills API, and the Files API is available on the Anthropic developer platform. The GA launch is live now for all Claude Platform accounts, with no migration required from existing beta configurations.