AWS has launched a public preview allowing AI agents to operate legacy desktop applications through Amazon WorkSpaces virtual desktops, using computer vision and input simulation instead of APIs. Agents authenticate via IAM, connect to isolated WorkSpaces instances, and interact with applications by taking screenshots, clicking, and typing — no application modification required. The solution supports any MCP-compatible agent framework (LangChain, CrewAI, Strands Agents) and inherits enterprise security controls including CloudTrail audit logs and CloudWatch observability. The key tradeoff is cost: Reflex benchmarks show vision agents consume 45x more tokens than API-based agents, but for organizations running legacy ERP or thick-client apps without APIs, this may still be cheaper than multi-year modernization projects. Microsoft is pursuing a similar approach with Windows 365 for AI agents.
Sort: