Trade-offs

What AI takes away.

Most writing about AI is about what you gain. This page is about three things it costs a law firm, and what I'd do about each. None of them is a reason not to use AI.

01Trade-off One

It can crowd out conversation inside the firm.

If memos, client updates, and the questions you'd normally ask a colleague all go through a chat window instead, the firm loses the exchanges where trust is built and judgment is passed on: the question in the hallway, the call where you hear a client hesitate, the partner explaining why a draft was wrong.

How I'd manage it Decide in advance which conversations stay human (client calls, difficult news, supervision, anything where reading the room matters) and make that firm policy. Use AI for the preparation and paperwork around those conversations, not for the conversations themselves.
02Trade-off Two

It makes it easy to stop checking.

When output is fluent and mostly correct, people review it less. Studies of knowledge workers have found the same pattern: the more people trust a tool, the less they scrutinize what it produces.

It also matters whether people understand roughly how the tool reached its answer. Without that, you can't predict where it's likely to be wrong, so you either check everything with equal effort, which nobody keeps up, or check nothing.

How I'd manage it Build verification into the workflow instead of relying on willpower; the CanLII tool does this by checking citations against the real database as the work happens. And train people on how these systems produce an answer, so they know which outputs to distrust.
03Trade-off Three

It can stop junior lawyers from learning the job.

The work AI handles best (document review, first drafts, research memos) is the same work that used to teach junior lawyers their judgment. Take it away and in a few years you have lawyers supervising AI output without the experience to know when it's wrong.

How I'd manage it Keep juniors doing the reasoning even where the tool could do the task: have them draft first and use AI as a second pair of eyes, or have them critique its output.