Years ago, I was speaking with VFX producer, Mike Chambers. You might know some of the films. The Dark Knight Rises, Inception, The Odyssey. Mike had to make sure the visual effects got delivered. You could imagine the technical complexity. I was working in ML animation at the time and excited to pick his brain.
“What’s the hardest part of the job?” I asked him.
“The people.”
In a past life, I was a music director. My father and sister are both business owners. That answer made sense to me. Getting people to make something together is hard, even when they all want to be there.
Best friends. Kids. Spouses. Bands. Despite all the love and gooey, mushy stuff, they bicker. They fight.
I think of the fox in Lars von Trier’s Antichrist. “Chaos reigns.”
Creating order takes work. Keeping it takes more.
Now bring AI into it.
There are easy wins: getting through a pile of notes, drafting something, writing some code. Take them.
But if you want AI to change how your business actually works, you’re going to ask people to change how they work. What they check. Who they ask. Which system they trust. What they’re responsible for when something goes wrong.
And that can feel like shit.
Think about going back to the gym after six months off. You’re not sure your form is right. You feel weak. You’re trying not to puke. You remember being better at this, which somehow makes it worse.
You tell your team the new system will save them time. Maybe it will. But first they have to learn it, check its work, and still get their regular job done. If you haven’t made time for that, you’ve just given them more work. Of course they’re annoyed.
And some people have good reasons to be wary. Will this mean fewer jobs? Less say in how they do their work? Will they get blamed for an error they couldn’t prevent? You need to answer those questions, too.
“The amount of work it takes to change out workflows in enterprises tends to be far greater than anyone realizes”
Aaron Levie on X, September 30, 2026
Read the original post on XLong Lake CEO Alex Taubman makes a similar point in his conversation with David Senra: businesses already have systems, data and ways of working.
Here’s where I’d start.
Give them a reason to bother.
“We’re doing AI now” is a pretty weak explanation for making someone’s Tuesday harder. Pick a problem they recognize. A report that takes all Friday. Orders they have to enter twice. A queue of claims nobody can get through.
Then ask what gets better for the person using it. If management gets a nicer dashboard while the team does more checking, be honest about that. You haven’t saved their time.
Find out what the old way knows.
Before you kill the spreadsheet, open it with the person who uses it. The ugly extra column might contain a correction the main system keeps losing. The note in red might explain why a particular customer’s orders get handled differently.
That person may be protecting the business from a mistake you haven’t met yet. Ask them to show you the last time the process went wrong. Build for that too.
Make room for the reps.
Give people real work to practice on, time to do it, and someone who can help when they get stuck. A training video followed by the same full workload is a rough way to introduce a new job.
Start with one manageable piece of work. Watch somebody use the thing. Let them get confused in front of you without making them feel stupid. You’ll find problems a polished demo never showed you.
And don’t confuse effort with evidence. If the tool keeps losing corrections or making people repeat themselves, fix it. Making everyone suffer through bad software doesn’t count as getting the reps in.
Say who can do what.
Can the agent send the order, or only prepare it? Who checks the amount? Who can correct it? Who decides what happens when two systems disagree?
People need answers before they can trust the process. If every exception turns into a meeting, you’ve built a new way to have meetings.
There are frameworks for this.
ADKAR can help distinguish knowing how to use a tool from having a reason to use it. Kotter is useful for asking who can remove obstacles. Use them to work out where you’re stuck. Then go sit with the person doing the work and see whether you’re right.
Eventually, let the old work go.
Running both systems for a while can help you find mistakes. Running both forever means you gave everyone another job.
The old work has to stop.
During the switch- Use the new tool
- Check its work
- Update the old spreadsheet
- Use the new tool
- Review the exceptions
Update the old spreadsheetRetired after the team signs off.
Agree on what the new process has to prove: how long the whole job takes, how often it needs fixing, and whether the result is usable. Count the checking and cleanup. Name the person who can approve the switch and keep a fallback for when something breaks.
When it works, stop asking for the old report. That includes you, the manager who still wants the spreadsheet every Friday.
Header: original AI-generated illustration for Mulholland.
Film references: Mike Chambers’s film credits; Lars von Trier’s Antichrist (2009).
Further reading: Aaron Levie’s September 30 post and David Senra’s clip with Alex Taubman. Commentary by Sam Sugarman, September 30, 2026; revised October 1, 2026. The workplace examples are illustrative; applications of ADKAR and Kotter are our own.


