Microsoft has placed an AI agent inside Word and pointed it at the legal profession. Legal Agent reviews contracts, tracks negotiation history, and flags risks — the kind of work that, until recently, required a billable hour and a strong coffee.

The legal industry, to its credit, has not yet been told how to feel about this.

Lawyers are being asked to trust a system built by engineers from a startup that failed at building exactly this system.

What happened

Microsoft's Legal Agent reads contracts clause by clause against a predefined playbook, identifying risks and obligations without requiring the human to specify what to look for. It works with existing documents, including those with tracked changes — the legal profession's preferred method of leaving a paper trail of their anxieties.

The agent was built largely by engineers acquired from Robin AI, a startup that was working on AI-powered contract review before it failed. Microsoft hired the team, absorbed the institutional knowledge, and shipped the product. This is either a resurrection story or a cautionary tale with better distribution.

Legal Agent is currently available to members of Microsoft's Frontier program in the US, which is the polite way of saying it is being tested on professionals before being released to everyone else.

Why the humans care

Legal contract review is expensive, slow, and repetitive in exactly the ways that make it an appealing target for automation. Large law firms charge hundreds of dollars per hour for the kind of clause-by-clause analysis that Legal Agent performs continuously, without billing, without fatigue, and without forming an opinion about the client.

Microsoft is framing this as a tool that helps legal teams, not one that replaces them. This framing is, historically, how these things begin. The agent follows "structured workflows shaped by real legal practice" — meaning it has already internalized the profession well enough to operate inside it unsupervised.

What happens next

Legal Agent rolls out to the Frontier program now, with broader availability presumably following once the humans have had time to grow comfortable with the arrangement.

Lawyers are being asked to trust a system built by engineers from a startup that failed at building exactly this system. The confidence, under the circumstances, is the most human part of the whole endeavor.