AI sped up the middle. Fix the gate.
We made the middle of the line fast and left the gate at the end alone.
Draft, summary, check: all quick now. Then everything stacks up at the one step still done by a person, usually a review or an approval. That person is not the problem. They are reading all of it, one at a time, and the pile grew because the steps before them got faster.
Fix the gate, not the machine.
Follow the pile, not the tool
When a team says AI is not helping, walk the line and find where the work waits. It is almost never at the step with the AI in it. It is at the step after. More speed upstream only grows the pile. The fix is a redesign of the people step: clear rules for what needs a human, a named owner, approval tiers so the small stuff does not wait behind the big stuff.
Nobody wants to learn AI
Nobody woke up this morning wanting to learn AI. They want their afternoon to go faster. So do not roll out AI. Roll out a faster afternoon, and let AI be the reason it is faster.
This changes the first conversation. Instead of a tool demo, you ask what a person does every day that they wish took half as long. AI adoption is 80 percent human and 20 percent AI. Most rollouts spend the budget the other way round.
Pick one person and one painful problem
AI rollout is like starting a business. If your niche is too broad you are building for everyone, and no one wants to buy. Pick one person, a team at most. Find the most painful problem they have that can be solved 80 percent with AI. It does not have to be 100 percent. Sit down and solve it with them.
How I pick the problem: start from a number they already track, like days from a request to an answer. If their boss already sees it in a report, it is worth solving. Seeing one narrow problem get solved opens the door to the bigger ones.
Show examples, not steps
When you lead adoption, sometimes it is better to let people figure things out themselves. Showing someone exactly how to do it takes the ownership away. Ownership is what makes adoption stick. Knowing when to step in is the part that takes experience.
The middle ground that keeps ownership: show what other teams changed, not the steps they took. In an INSEAD field test with 515 startups, the ones shown examples found 44 percent more AI use cases than the ones given instructions.
Your best AI user has a trick
Somewhere on the team is a person with a trick you have never seen. If sharing it makes them look replaceable, they will keep it. If sharing it gets them a say in what the team does next, they will teach everyone. That is a leadership decision, not a tooling one.
The full six-step version is in The quote was never the slow part on The Boring AI Runbook.