Microsoft just built an AI that reads contracts the way lawyers actually work, not the way ChatGPT thinks they work.
The Summary
- Microsoft is launching Legal Agent, an AI agent inside Word designed specifically for legal teams to handle document edits, negotiation history, and contract review
- Unlike general AI models that interpret commands, this agent follows structured workflows based on real legal practice, reviewing contracts clause by clause against playbooks
- The agent works with existing tracked changes in documents, meaning it plugs into how legal teams already work rather than forcing new tools
The Signal
This is the kind of vertical AI agent that actually matters. Not a chatbot that can summarize a contract. Not a generative model that hallucinates case law. Microsoft built Legal Agent to follow the actual workflow of legal review: clause-by-clause comparison against established playbooks, tracking changes across negotiation rounds, and working inside the document format lawyers already use.
The key phrase from Sumit Chauhan, Microsoft's corporate VP: "structured workflows shaped by real legal practice." This is the difference between AI tools that lawyers might use and AI agents that do legal work. General language models are pattern-matching machines trained on the internet. This thing is trained on how contracts actually get negotiated.
"Instead of relying on general AI models to interpret commands, the agent follows structured workflows shaped by real legal practice."
Legal work is high-stakes, detail-oriented, and expensive. It's also extremely structured once you know what you're looking at. Every NDA follows a pattern. Every employment agreement has the same sections. Every M&A deal has similar sticking points. That structure is exactly what makes legal work automatable, but only if the agent understands the domain.
The fact that Legal Agent works with tracked changes is not a small detail. Lawyers don't start from scratch. They redline. They iterate. They compare version 7 to version 12 to see what the other side tried to slip in. If your AI can't handle that workflow, it's a toy. If it can, it's a colleague.
Here's what this tells us about where agents are headed:
- Domain-specific beats general-purpose when real work is on the line
- Workflow integration matters more than flashy capabilities
- The agents that win will be the ones that fit into existing tools, not the ones that demand you switch platforms
Microsoft is making a bet that the future of knowledge work isn't "talk to an AI and it does magic." It's "this specific agent knows how to do this specific job inside the tool you already use." That's a much less sexy pitch. It's also probably correct.
The Implication
Watch how fast other professional service verticals get their own agents. Accounting, consulting, HR, procurement. Anywhere the work is structured, repeatable, and expensive, there's a vertical agent being built right now. The race isn't to build the smartest general AI. It's to build the agent that best understands how accountants actually close books or how HR teams actually process onboarding.
For knowledge workers: if your job is highly structured and you do the same type of task repeatedly, your agent is coming. The question is whether you learn to manage it or whether it learns to replace you.