Skip the training-first approach. The resistance most people are worried about isn’t fear of the technology, it’s fear of looking slow or replaceable in front of your peers, and no amount of prompt engineering workshops touches that. The honest answer to that fear isn’t a workshop. It’s that we don’t replace people, we remove the step that never should have been manual in the first place. The person stays. The repetitive part disappears.
Three things actually make that real, and they attack the fear from three different angles: what gets automated first, who gets to judge whether it’s any good, and whether leadership is honest when something fails.
This assumes the workflow itself is ready. If you’re not sure, run the readiness test for AI automation first, because none of this works on a process nobody has documented.
What to Automate First When Your Team Is Skeptical of AI
Make the first thing you automate something people genuinely hate. Not the interesting part of their job, but the part they complain about at lunch. You want the team’s first experience of AI agent automation to feel like relief rather than evaluation.
That choice of what to automate first is the whole game. Get it right, and the team’s first data point about AI is that it took something off their plate. Get it wrong, pick the interesting part of the job instead, and the first data point becomes “this is coming for what I’m good at.”
Put Your Subject Matter Experts in the Reviewer Seat, Not the User Seat
Once that first win lands, the second move is about who gets to judge the work, not just what work gets automated. Put your experts in the reviewer seat rather than the user seat. Ask them to grade the agent’s output and tell you where it’s wrong. That reframes the whole thing, because now the AI is the junior employee and they’re the authority, which is both truer and much easier to accept.
This also gives you a way to build up and train the system with your team’s knowledge. Their corrections are the highest quality training material you will ever get, because it’s coming from the person who actually knows what right looks like.
Don’t let those corrections just disappear into a spreadsheet somewhere. How to build a knowledge base for AI agents walks through where they should actually go. And if you’re at a professional services firm, how firms encode expertise before a competitor does makes basically the same case for your world.
Kill a Failing AI Pilot in Public, and Do It Fast
The other move that builds real trust is publicly killing something that isn’t working, and doing it quickly. Teams watch closely to see whether leadership will admit an AI project failed or if they keep pretending. If you kill one honestly, everyone believes you about the next one. This is one of the reasons why AI pilots stall before production so often, quietly, without anyone formally calling it.
Put together, none of these three moves are really about technology. What you automate first, who’s trusted to grade it, and whether leadership tells the truth when something doesn’t work are all questions about people, not models. Get those right and the technology stops being the hard part.
Find the First Workflow Worth Automating
If you’re not sure which process would actually feel like relief to your team instead of another evaluation, that’s worth talking through before anything gets built. We can walk your workflow and tell you where the judgment lives, and what encoding that judgment into a rules engine would actually involve. Or if you’d rather start smaller, take the free AI readiness assessment first.
