Part II: AI Agents Are Entering the Legal Workflow. Now Comes the Hard Part.

The conversation has changed. We’re talking less about whether lawyers should use AI and much more about what happens once AI actually becomes part of the work.

What does that mean for governance? For billing? For the client-counselor relationship? And, increasingly, what does it mean for how legal work itself is structured?

One session that particularly caught my attention was Whose Legal AI Workflow Is It Anyway?, which took real legal use cases and broke down how agent-based workflows could be built around them including what AI should do, where lawyers need to stay involved and where approval and escalation points belong.

Another session on my list today is Change Management for Legal Technology Transformation, which gets to the other half of this problem: what happens to the organization when the technology begins changing the work?

Together, those topics feel like a pretty good snapshot of where legal AI is right now.

A few years ago, the question was: Can AI do legal work? Then it became: How do we get lawyers to use it? Now we’re moving toward: What does legal work look like when AI is actually inside the workflow?

And there is another reason this is becoming urgent.

AI is moving onto the desktop. Microsoft, for example, now has a Legal Agent inside Word that can review agreements, generate redlines and analyze contracts against an organization’s playbook.

That is a very different experience from opening ChatGPT and asking it a question.

The AI is starting to sit where the work already happens. And as agents become more capable, they will increasingly handle not just individual tasks, but pieces of an entire workflow.

That changes the management problem.

Take contract review.

A traditional process might involve intake, initial review, issue identification, drafting proposed changes, attorney review, negotiation and execution.

Now imagine an agent can conduct the initial review, identify deviations from the playbook and propose the redline before a lawyer ever touches the document.

Organizations suddenly have to answer some very practical questions. What is the agent allowed to do? What requires human review? Who is responsible for the output? What information can it access? What happens when something goes wrong?

And for a law firm: What do you bill the client when work that once took several hours can now be completed much faster?

But there is another question that may be even more important: If an agent can collapse three steps of a seven-step process, why do we still have a seven-step process?

That is where we move from AI implementation to organizational redesign. And the distinction matters.

McKinsey’s latest research found that nearly three-quarters of AI high performers are fundamentally redesigning workflows around AI, compared with only about one-quarter of other organizations. That makes sense.

Putting AI into an inefficient workflow does not suddenly make the workflow good.

Sometimes it just automates the inefficiency. And that is why I think the change-management challenge is about to get harder, not easier. Organizations will not just have to teach people how to use AI.

They will need to rethink policies around AI and communicate those expectations to employees, suppliers and clients. They will have to teach people how to work alongside agents.

Managers may increasingly be managing work performed by both people and AI. Legal and IT will have to make decisions about access, oversight, governance and accountability.

Leaders will have to think differently about training, roles and performance when parts of the work can be done dramatically faster — or potentially no longer need to be done by a person at all.

And lawyers themselves will have to become very good at knowing where human judgment actually matters and where it does not.

That is a much bigger organizational shift than rolling out another piece of software.

With one more day left at ILTACON, the question I’m most interested in isn’t: What can this AI do? It is: If it works, what are we actually prepared to change? The workflow? The roles? How people are trained? What gets measured? What gets rewarded? How lawyers spend their time?

Because that may ultimately determine which organizations get real value from AI — and which simply add another tool to the way they have always worked.