Software is now a coworker. A partnership between Microsoft and Workday has created the first formal system for managing AI agents as digital employees. They are given unique identities, tracked in HR systems, and measured on performance. This shift from tool to worker raises urgent questions of control, security, and who is responsible when the algorithm makes a mistake.
This is Modra, just north of Bratislava. The vineyards are old, but the work is changing.
In an office tower miles away, a new worker starts its first day. It has no desk and no family. It has a job. It has an identification number. And it has a permanent record in the company’s human resources system. This worker is not a person. It is an autonomous AI agent.
An ID and a File
This is not a forecast. The architecture is already built. On September 16, the enterprise software firm Workday announced it was connecting its new Agent System of Record, or ASOR, to Microsoft’s Entra Agent ID. The partnership gives a piece of software a formal identity. It gives it a place on the organizational chart.
For years, automation was a tool, like a hammer or a spreadsheet. Now, it is becoming a digital employee. The shift is quiet but absolute. Microsoft’s system, announced in May, acts as a digital birth certificate, issuing each agent a unique, auditable identity. Workday’s system is the HR file, tracking the agent’s role, permissions, and performance alongside its human colleagues.
The Ledger and the Lock
The purpose is control. In a world of automated systems, the question “Who did that?” can be hard to answer. This new model provides a ledger. The PagerDuty survey firm reports that companies are already running multiple agents. Citigroup is piloting them. The agent is moving from the lab to the front office.
But identity is a fortress, and every fortress can be tested. On September 22, security researchers reported a critical flaw in Microsoft’s broader Entra ID system. The flaw was patched weeks earlier, but it underscores the stakes. A compromised human identity is a problem. A compromised agent identity, with access to core systems, is a catastrophe.
The Unwritten Rules
An agent is not just hired; it is measured. Workday’s own guidance outlines Key Performance Indicators—KPIs—for its digital workers. They are graded on accuracy, on cycle time, on the number of times they require human intervention. They can even be cited for safety violations.
The system is live. The digital worker is being provisioned. This leaves the most difficult questions unanswered. When an agent makes a costly error, who is liable? The developer, the platform, or the human manager who deployed it? No one has written that policy yet.
The agent-as-employee is no longer a concept. It is a product with a stock-keeping unit. The technical framework is in place; the corporate governance for a hybrid human-and-agent workforce has yet to be invented.