Your first version with AI loses to doing it by hand

A product manager is hired to build the thing. An ops person is hired to do the ops.

Incentives drive a lot of behavior. That's something few are accounting for when they tell operations teams to start building. When a product manager's first attempt fails, that's the job. Failing is how you improve. When an ops person's first attempt fails, the orders didn't ship and the invoices didn't post, and the afternoon looks wasted to them and to their manager.

And the first attempts will lose. Your first AI workflow takes longer than the task it replaces. You spend an hour setting up something that would have taken you fifteen minutes by hand. The second version isn't much better. It usually takes three or four iterations before the workflow is faster than your fingers, and by then most people have stopped.

I see this often. Someone gets excited, tries once on a Friday afternoon, hits a snag, and is back in the spreadsheet on Monday. Not because they didn’t want to try improving it. But because they were hired to process spreadsheets, not to build the process, and they hit a deadline.

It's not laziness and it's not resistance. It's a rational response to how their time is measured.

That's why our builder program takes 50% of someone's time for 90 days, and why leadership had to sign off on it. When we first proposed this, some departments pushed back and asked for 10% of their builder’s team. But we knew this wouldn’t be enough to account for the losing iterations, so people wouldn’t have time to actually change how they work.

We hope this also shifts the narrative for builders after the program. By having builders across teams, we can decentralize experts and organically share lessons like this. Soon, the incentives of ops teams will change, so we don’t need a program like this.

If you want people who do the work to also build the tools, budget for the versions that don't work yet.