AWS is introducing new features to Amazon WorkSpaces that allow AI agents to independently run desktop applications. With this, the company is targeting organizations that don’t have modern APIs and still rely heavily on legacy software and applications.
This enhancement allows AI agents to operate within the same virtual desktop environment that organizations already use for their employees. According to AWS, this allows you to automate existing business processes without first modernizing your applications or migrating them to a new platform.
Many large organizations still run business-critical processes in outdated software environments. According to figures cited by AWS from Gartner research, the majority of these applications do not have programmable interfaces. This makes it difficult for modern AI systems and automation to access it.
In AWS, agents work through the application’s user interface instead of directly integrating APIs. AI agents can access managed WorkSpaces desktops, click, type, scroll, and analyze screen information via screenshots. In reality, the agent functions similar to a human user on a desktop.
This solution leverages existing AWS services for authentication and logging. AI agents authenticate through AWS Identity and Access Management, and activity is logged through AWS CloudTrail and Amazon CloudWatch. The agent works within your existing WorkSpaces environment, so your security and compliance settings remain in place.
Additionally, the service supports Model Context Protocol, an open standard for AI agent integration. This allows organizations to use a variety of agent frameworks such as LangChain, CrewAI, and Strands Agent.
AWS demonstrated the functionality using an example where an AI agent processes repeat prescriptions independently within a pharmacy application without any API integration. Agents retrieve patient data, select medications, and complete orders all through a desktop interface. According to AWS, no changes were required to the application.
Criticism of the efficiency of AI agents
But this approach also has drawbacks, The Register reports. So-called computer-enabled agents can be significantly more expensive and less efficient than traditional API integrations, according to research from AI company Reflex. Agents are operated via screenshots and computer vision, which requires more AI interaction to perform simple tasks. According to Reflex, a simple interaction with the user interface can already require hundreds of thousands of tokens. Therefore, the company concludes that AI agents that operate through a desktop interface are still structurally more expensive than software that communicates directly through an API.
AWS, on the other hand, is specifically aimed at helping organizations that lack modern APIs. Therefore, the WorkSpaces approach effectively serves as an alternative to costly traditional software modernization projects. Instead of modifying your applications, AWS allows your agents to operate through the same user interface that your employees already use.
AWS also emphasizes the security benefits of an isolated virtual desktop environment. AI agents run within separate WorkSpaces instances, so they don’t require direct access to local endpoints or your corporate network. This infrastructure is also suitable for short-term agent workloads because WorkSpaces can be started temporarily and then shut down.
AWS is not alone in this development. Microsoft is also working on support for AI agents within the Windows 365 environment. This has created a new category of cloud desktop services, where AI systems not only call applications via APIs, but actually interact with the software through a user interface.
