Last night I was talking with an attorney at another firm about a trend I believe many transactional lawyers are seeing: clients are using AI not only to understand agreements, but also to question the advice they receive.

His concern is easy to understand. Whether a client uses a tool to pressure-test our advice or simply to understand a provision, the underlying change is the same. The client now has a second reader.

Picture a client who reads our advice, runs it through an AI tool, and comes back with a provision the tool says deserves attention. The unspoken question is whether we missed something.

The most common response is that the AI model lacked context, and that is often true. A model can identify something worth discussing without knowing whether the provision is actually a problem. It does not know why an attorney accepted a compressed timeline, or why they did not fight harder on remedies. Those are usually not oversights. They are trades, made because of how the parties have dealt with each other before, and because the client wanted more. But that answer is only part of the story.

The trouble with that answer is that the context gap is closing. Deal history, prior agreements, the term sheet, the email traffic, all of it is context, and will be increasingly available to the model. If our response is that the machine does not know what we know, we have a short runway.

The durable difference is not what we knew. It is that we decided, and that we are answerable for the decision. A model produces every issue theoretically available. An attorney decides which three are worth spending relationship capital on, in a live negotiation, with a client’s deal and a firm’s name behind the call. That is a different act.

The late Clayton Christensen, a Harvard Business School professor who spent his career studying why capable, well-run companies get displaced by cheaper competitors, framed this better than I can. Customers do not buy products, he argued. They hire them to do a job. Our clients do not hire us to produce a marked up contract. The markup is the deliverable, not the job. The job is closer to the older word for what we do: counsel. Help me see whether this deal works for me in year three. Tell me which risks are worth carrying to get it closed, and say so plainly when I am paying for protection I might never use.

There is another Christensen lesson, from How Will You Measure Your Life, that worries me most. He observed that we allocate our time and energy toward whatever pays off soonest and measures most cleanly. In our business, that is the billable hour. Understanding a client’s business, their board, the pressure they are under, and what their concerns are pays off slowly, invisibly, and on no timesheet at all. It is the investment we defer. It is also the one thing a model cannot replace.

What is easy to miss is more ordinary. The advantage is not that an attorney reads the contract better than the model. The advantage is understanding the people behind it. That knowledge comes from paying attention to people, which is the part of the job that is difficult to measure and can receive less attention than the technical work directly reflected in a deliverable or timesheet.

So, when anyone on a deal team flags a provision, I want us to get past whether it is market and understand why it is there and what was traded away to get it. I don’t want to just spot the issue. I want to tell the client whether the bargain still serves them. That is not a technology strategy. It is counsel, which is what they hired us to do.

As AI makes technical work faster and more accessible, attorneys have an opportunity to devote more of their time to the work clients value most: exercising judgment, explaining tradeoffs, and helping them make better decisions.