Microsoft Copilot Studio just achieved a major milestone: computer-using agents are now generally available. After months in preview, this capability has matured into an enterprise-ready platform feature that lets AI agents interact directly with websites and desktop applications through their user interfaces—no APIs required.
This is a big deal for organizations that have been stuck between modern AI ambitions and legacy systems that lack integration hooks. Let me walk you through what's new, why it matters, and how you can start putting it to work.
What Are Computer-Using Agents?
Computer-using agents are AI agents that can "see" and interact with software the same way a human does—through buttons, forms, menus, and web pages. Instead of relying on APIs, webhooks, or brittle screen-scraping scripts, these agents use visual understanding and reasoning to navigate applications directly.
Think about all the business processes that span systems without APIs:
- •Legacy ERP modules that haven't been modernized
- •Vendor portals that change their layout quarterly
- •Internal tools built decades ago that still run critical workflows
- •Government or partner systems that only offer web interfaces
Traditionally, automating these processes meant either expensive modernization projects or fragile RPA (Robotic Process Automation) bots that break every time a UI changes. Computer-using agents change the equation by bringing adaptability to the automation stack.
What's New in General Availability
The GA release brings several enterprise-hardening improvements beyond what was available in preview:
Secure Credential Management Organizations can now manage credentials more securely for agents that need to log into applications. This includes support for credential stores and rotation policies that IT teams need before they'll sign off on production deployments.
Model Choice for Different Scenarios Not every automation task needs the same level of reasoning power. The GA release lets you choose models best suited for different scenarios—lighter models for straightforward form-filling, more capable models for complex multi-step workflows with decision points.
Resilient UI Adaptation This might be the most practically important improvement. The agents can now adapt when interfaces change—buttons move, labels get renamed, new fields appear. Instead of breaking immediately like traditional RPA, the agent can reason about the new layout and continue working. That's the difference between maintenance-heavy bots and truly adaptive automation.
Embedded in Multi-Step Workflows (Preview) A particularly powerful new capability lets you embed computer-using agents directly into Copilot Studio workflows. This means you can combine API-based actions, approvals, business logic, and adaptive UI interactions all within the same automation flow. Structure where you need it, adaptability where you don't have APIs.
A Real-World Example: Graebel
[Graebel](https://www.graebel.com/), a global talent mobility company, illustrates the power of this approach beautifully. They process thousands of employee relocation requests annually, most arriving as unstructured emails with unique instructions and attachments.
Their proprietary Global Connect platform lacked API support, so traditional automation couldn't handle the variability. Working with GET AI and Microsoft, they built the Graebel Service Order Agent in Copilot Studio using computer use capabilities. The agent can:
- •Interpret incoming emails with all their unstructured complexity
- •Validate requests against business rules
- •Operate Global Connect directly through the UI
- •Escalate exceptions through workflows when needed
The results? Reduced manual effort, faster turnaround times, more consistent data quality, and a repeatable blueprint for bringing intelligent automation to over 30 relocation service categories.
As Matt Brownlee, Graebel's Chief Revenue Officer, put it: *"By adopting Microsoft Copilot Studio and AI agents, we've moved beyond traditional automation to a more intelligent, scalable operating model."*
The Bigger Picture: Connected, Adaptive Automation
Computer-using agents don't exist in isolation. Microsoft has been building toward a more connected automation ecosystem, and several complementary announcements show the full vision:
Redesigned Workflows Experience A new visual designer for building and orchestrating agentic automation end-to-end. You can design workflows on a unified canvas, incorporating AI-powered actions like classification and content generation alongside traditional automation steps.
Crucially, you can add agent nodes to workflows—injecting adaptive AI intelligence at decision points that can't be captured in simple if-then logic.
Agent-to-Agent Communication With A2A (agent-to-agent) communication now generally available, specialized agents can exchange information and delegate tasks across systems. Your computer-using agent can hand off to an API-based agent when it reaches a system with proper integration hooks, and vice versa.
Work IQ Interoperability New REST API and CLI capabilities for Work IQ, plus support for remote MCP (Model Context Protocol) servers, provide standardized ways to connect agents with tools and enterprise resources without one-off integrations.
Getting Started
If you're ready to explore computer-using agents, here's a practical starting path:
1. Identify a candidate process — Look for workflows that currently require manual UI interaction, especially ones that are repetitive but too variable for traditional RPA.
2. Start in Copilot Studio — The [computer use documentation](https://learn.microsoft.com/en-us/microsoft-copilot-studio/computer-use) walks through building your first agent.
3. Plan for governance — The GA release includes enterprise controls, but you'll still want to think about credential management, approval workflows, and monitoring before production deployment.
4. Consider embedding in workflows — For processes that span both modern and legacy systems, experiment with the new workflows experience to combine API-based and UI-based automation.
Looking Ahead
Microsoft Build is happening this week (June 2-3, 2026), and the agentic automation track is packed with sessions covering everything from multi-agent orchestration to distributed agentic apps running from edge to cloud. If you're investing in this space, it's worth catching the recordings.
The trajectory is clear: Microsoft is building toward a world where AI agents can work across the full landscape of enterprise software—modern APIs where available, direct UI interaction where necessary, and intelligent orchestration tying it all together. Computer-using agents reaching GA is a major step on that path.
*What's your take? Are you already experimenting with computer-using agents, or still trying to identify the right use case? I'd love to hear about your automation challenges in the comments.*