Julia runs operations at a 40-person law firm. She's the person the managing partners turn to when something new needs to actually work. New case management system? Julia figured it out. Remote work transition in 2020? Julia built the playbook. Office relocation? Julia.
So when the managing partner came back from a conference last fall and said "we need to be doing something with AI," the assignment landed on Julia's desk. Not because she's technical. Because she's the one who knows how work actually gets done.
And so, she did what she's always done. She researched the options, got licenses for a platform that made sense for legal work, and organized a training session with the vendor. She made sure every attorney and paralegal had access and a basic understanding of what the tool could do. She sent a follow-up email with resources and even set up office hours for questions.
That was four months ago. Today, practically nobody is using it.
A few associates dabble, and one paralegal uses it for first drafts of routine correspondence. But the partners who pushed for this aren't touching it themselves, and the firm's actual workflows haven't changed at all. The licenses are active, the tool is available, and nothing is different.
Julia has run this play before and it has always worked. Get the tool, train the people, support the transition, measure adoption. This is how competent operational leaders have handled every new technology for the past fifteen years. She executed it well. She knows she executed it well.
So why does she feel like she's failing?
If you've been in Julia's position, you already know the particular discomfort of this moment. It's not just that the initiative stalled. It's that her professional identity is built on being the person who figures things out, who translates ambiguous executive mandates into operational reality. She's always had a plan. And right now, she doesn't have one. Not because she hasn't tried, but because the plan she had (the one that's always worked) produced nothing. And she can't see what the next plan should be.
That gap between "I'm good at this" and "this isn't working" is disorienting in a way that a normal project setback isn't. A normal setback has a cause you can identify and fix. This one feels like the ground shifted and nobody told her.
Meanwhile, the noise keeps coming. The managing partner forwards articles about firms using AI to do in minutes what used to take hours. A junior associate mentions a competitor that's apparently "way ahead." A consultant calls offering an AI readiness assessment. An attorney at a CLE event tells Julia that her firm needs an AI policy before someone uses client data in a prompt. Everyone is saying "do something," and nobody is saying what.
She's the person who always has the answer. Right now she doesn't have one. And the dissonance between those two things is louder than any of the external noise.
"She's the person who always has the answer. Right now she doesn't have one. And the dissonance between those two things is louder than any of the external noise."
Here's what I want Julia to know, and what I want you to know if you recognize yourself in this story.
The playbook didn't fail because you executed it poorly. It failed because what you were handed wasn't actually a deployment problem. "Go make AI happen" sounds like a mandate with a clear shape: select a tool, roll it out, train people, drive adoption. That's a deployment. And Julia is great at deployments.
But AI adoption at an organization isn't a deployment. The tool is the easy part. The hard part is everything around it.
It's the managing partner who championed the initiative but won't use the tool himself. He pushed for this because he heard about it at a conference and knew the firm needed to move. But actually sitting down and typing a prompt in front of people? That would mean being a beginner again, visibly, in a firm where he's spent twenty years being the authority in the room. So he delegates the enthusiasm downward and never models the behavior himself.
It's the senior attorney who's quietly terrified that this technology devalues her. She's spent thirty years building the kind of pattern recognition that makes her invaluable to clients. She can read a contract and spot the risk in paragraph fourteen because she's seen what happens when you miss it. The suggestion that a machine can do some version of that in seconds doesn't feel like progress to her. It feels like erasure. Nobody has told her otherwise, so she assumes the worst and keeps her distance.
It's the paralegals who would actually benefit most but don't feel safe experimenting. Nobody has said clearly what's okay to use it for and what isn't. Can they put client information into a prompt? Will they get in trouble if the output is wrong? Is this supposed to replace part of their job? In the absence of clear answers, the rational move is to do nothing. So they do nothing.
Those aren't training problems. Those are problems of trust, clarity, identity, and permission. And no amount of licenses or lunch-and-learns will touch them.
The assignment Julia received ("we need to be doing something with AI") was a wish masquerading as a strategy. It had no scope, no success criteria, no acknowledgment that the real barriers are human rather than technical. It was the organizational equivalent of declaring a destination without funding the trip.
"The assignment was a wish masquerading as a strategy. It had no scope, no success criteria, no acknowledgment that the real barriers are human rather than technical."
If this is where you are, the most useful thing I can tell you is this: the fact that your usual playbook isn't working is not evidence that you're behind. It's evidence that the problem is shaped differently than you were told.
You're not failing at the assignment. The assignment was incomplete. And recognizing that is the first step toward doing something that might actually work, which starts with understanding that you're not solving a technology problem. You're solving a change problem that happens to involve technology. Those require different tools, different timelines, and a fundamentally different kind of leadership support.
You might not be able to fix all of that from your position. But you can stop running a playbook that was designed for a different problem. And you can stop blaming yourself for the fact that it didn't work.
That clarity, on its own, is worth more than another vendor demo.
Break a Pencil,
Michael
P.S. If you know a Julia (or you are one), I'd genuinely like to hear what you've tried and where you got stuck. Reply to this email. I read every response, and these conversations are shaping how I think about this problem.