Modernize your workflows: Amazon WorkSpaces now gives AI agents their own desktop (preview)

The gap between legacy systems and modern AI has always been one of the trickiest problems in enterprise automation. You’ve got decades-old desktop applications that run critical business processes, but they were never designed to talk to AI agents. Until now, bridging that gap meant either ripping and replacing the entire system or building custom integrations—both expensive and risky propositions. AWS is tackling this differently with a preview feature that lets AI agents operate desktop applications directly through Amazon WorkSpaces, the managed virtual desktop service. Instead of modernizing your backend, you’re giving the AI agent its own desktop environment to interact with applications the way a human would.

Here’s how it actually works: AWS integrates the Model Context Protocol (MCP) with WorkSpaces, allowing AI agents to connect to a virtual desktop instance with full IAM authentication and security controls. The agent can see what’s on screen through computer vision capabilities, click buttons, fill forms, and navigate applications just as a human operator would. This is important because many legacy systems—insurance claim management tools, ERP interfaces built in the 1990s, proprietary manufacturing software—don’t have APIs and never will. The agent operates through the GUI layer, which is something every application has. You set up the WorkSpaces instance with the applications you need, configure IAM roles to control what the agent can access, and let it run. The environment stays within your existing security framework, audit trails, and compliance boundaries.

Practically, this opens up automation possibilities that were previously out of reach. Consider a financial services company processing mortgage applications through three different legacy systems that don’t talk to each other. An AI agent could log into each system, extract data from one, input it into another, validate results in the third, and flag exceptions—all without human intervention and all within a contained, auditable environment. Or take a healthcare provider managing patient records across older hospital systems: the agent could gather information from multiple interfaces, consolidate it, and trigger downstream workflows. The key advantage over “traditional” robotic process automation is that this is cloud-native, uses modern IAM controls, scales through AWS infrastructure, and integrates with your existing GenAI tools and frameworks.

The preview phase is important to watch if you’re managing legacy desktop applications that have been automation bottlenecks. This isn’t a complete solution yet—you’ll want to test it with non-critical processes first, plan for edge cases where the agent might encounter unexpected UI changes, and think through how you’ll handle exceptions that require human judgment. But the foundation is solid: it removes the false choice between expensive modernization and accepting manual work, and it gives you a way to inject AI automation into workflows that were previously locked behind legacy UIs.