DAILY ISSUE · 2026-07-08 · AWS · AGENTIC AI

AWS Gives AI Agents a Desktop — A New Path for Legacy ERP and Mainframe Automation

Amazon Web Services has made Amazon WorkSpaces Applications generally available, opening a path for AI agents to securely access and operate desktop applications. ERP systems, mainframes, and legacy tools without APIs can now become targets for AI automation. This is not just a feature launch. It signals that one of the last practical barriers to enterprise AI transformation may be starting to fall.

What happened

AWS previewed Amazon WorkSpaces for AI Agents in May 2026, made it generally available on June 30, and formally briefed Korean media on July 7. Major Korean technology outlets including ZDNet Korea, CNET Korea, Byline Network, and TechDaily covered the announcement.

The core point is straightforward: AI agents can now access existing desktop applications through Amazon WorkSpaces, AWS’s managed cloud desktop environment. Agents interact with application UIs through point, click, and navigation patterns similar to a human operator. The connection runs through MCP, or Model Context Protocol, endpoints while using IAM authentication and existing enterprise governance controls.

According to AWS’s official positioning, the capability is designed to let AI agents securely operate legacy desktop applications such as ERP, CRM, mainframe, and proprietary tools that do not expose API access, without requiring application modernization. In practical terms, even an older COBOL-based mainframe screen or a legacy ERP system without an API can become something an AI agent can operate directly.

Why it matters now

The significance of this announcement sits on three levels.

First, it addresses the last mile of AI agent automation. Until now, most agent automation has been limited to modern SaaS tools and web services with APIs. The systems that carry core enterprise workflows — mainframes, ERP suites, SAP GUI environments, and internal administration tools — often lack APIs or have restricted access models. They have remained blind spots for AI automation. AWS is approaching that gap through desktop UI operation, combining computer vision and MCP so agents can “see” and manipulate the screen.

Second, it changes the pace of enterprise AI transformation. One of the biggest blockers for companies adopting AI agents has been the cost of making existing systems usable by AI. Adding APIs to legacy platforms or modernizing them can take months and large budgets. WorkSpaces for AI Agents lowers that barrier by allowing the agent to operate the UI directly. The underlying system does not need to be rewritten, which can materially improve the ROI profile of workflow automation.

Third, it expands what AI agents can actually do. Many existing agents are still concentrated in text-heavy tasks such as document summaries and email drafts. Desktop access moves the scope toward complex GUI-based business processes: entering purchase orders in an ERP system, updating customer records in a CRM, or retrieving data from a mainframe screen. The automation paradigm expands from “text processing” to “desktop operation.”

The Korean market context

The announcement has particular weight in Korea. Many Korean enterprises still operate IE-based ActiveX environments, older ERP deployments, and COBOL mainframe systems. Financial institutions, public-sector organizations, and large manufacturers remain especially dependent on legacy systems that are difficult to modernize.

That is why ZDNet Korea and CNET Korea treated the story as major technology news. Demand for enterprise AI transformation in Korea is high, but compatibility with legacy systems has been a persistent bottleneck. AWS’s announcement offers a technical path around that constraint.

There is still a governance question. Korean security frameworks and regulatory expectations, including those shaped by financial-security and ISMS-P requirements, do not yet provide a simple playbook for external agents operating desktops. AWS WorkSpaces provides IAM-based access control and audit logging, but each company will still need to design governance that fits the local regulatory environment. ZHS will return to this point in a deeper follow-up analysis.

The ZHS angle — what Hermes Agent is already doing

There is an interesting parallel with Hermes Agent, which ZHS already operates. Hermes Agent automates work in the terminal environment, using more than 34 practical tools for file management, code execution, data processing, web search, and related operations. If AWS WorkSpaces for AI Agents standardizes desktop GUI automation, Hermes Agent can be seen as standardizing CLI and terminal automation.

The difference between the two approaches is clear. AWS provides a managed cloud desktop so enterprises can run agents while preserving security and governance. Hermes Agent is open source, allowing individual developers and startups to install, modify, and extend it freely. AWS’s emphasis is enterprise security. Hermes Agent’s strength is flexibility and extensibility.

From the ZHS perspective, AWS’s announcement is another signal that AI agents using “human tools” are becoming an industry pattern. Standardizing the agent-to-desktop connection through MCP is conceptually aligned with how Hermes Agent connects tools in the CLI environment. CLI-based automation and GUI-based automation are beginning to converge around a common agent tooling model.

Competitors and alternatives

AWS’s entry makes the competitive landscape more complex.

Microsoft has already built an ecosystem where Copilot can operate across Windows and Office tools. However, Microsoft’s approach is optimized around its own product suite, including Microsoft 365 and Windows. It does not automatically extend the same automation coverage to third-party ERP systems or mainframes.

Google Cloud provides an agent-building platform through Vertex AI Agent Builder, but desktop UI automation is not as concretely packaged as AWS WorkSpaces.

The open-source ecosystem includes projects and approaches such as OpenClaw, Anthropic’s Computer Use API, and browser automation tools like Playwright and Puppeteer that offer partially similar capabilities. But these tend to focus on individual desktop or browser operations. They do not generally provide enterprise governance features such as centralized management, audit logging, and IAM integration at the level AWS can offer.

AWS’s clearest differentiator is that it works on top of infrastructure many enterprises already use. Companies do not need to build a separate agent infrastructure from scratch. They can reuse existing WorkSpaces environments and IAM policies, which is a decisive advantage in the enterprise market.

AWS’s announcement expands the practical boundary of what AI agents can do. Moving beyond API dependency and treating the desktop UI itself as an operating interface could become especially powerful in Korea, where legacy systems remain deeply embedded in enterprise operations.

ZHS will continue tracking this shift and plans to prepare a practical comparison between AWS WorkSpaces for AI Agents and open-source agent approaches based on our operating experience with Hermes Agent. The era of AI agents using human tools is no longer theoretical. This is one of its first concrete enterprise steps.