Every operator I know has the same instinct once revenue clears a threshold: automate everything. Wrong order. Automate the parts you’ve already mastered by hand. Everything else is guessing with better tooling.
I run three businesses on the same set of principles, different tech underneath. Wise STR Group needs guest communication, pricing, and turnover logistics running without me in the loop. Wise Media needs proposals moving to invoices without a human chasing them. The Web3 side needs settlement and compliance data that doesn’t blink. Same rule applies to all three: I don’t automate a decision until I’ve made that decision myself, by hand, enough times to know what correct looks like.
AI didn’t change this rule. It raised the cost of ignoring it. A model will automate your bad judgment as fast as your good judgment — faster, actually, because it never gets tired enough to notice something’s off. Hand a system a process you’ve never run yourself, and you’re not scaling expertise. You’re scaling ignorance at machine speed.
Marcus Aurelius kept a private notebook of instructions addressed to no one but himself. The one he kept returning to: know the thing before you assign the thing. He wasn’t writing about automation. The principle transfers anyway. If you can’t explain the judgment call, you can’t hand it to software.
Solomon put it plainer: a prudent man foresees the difficulty and hides himself, but the simple pass on and are punished. The automation graveyard is full of founders who skipped the foreseeing. They saw a competitor’s slick funnel and copied the output without doing the input — the hundred manual reps that taught the competitor which edge cases actually mattered.
Here’s the sequence I use across all three companies. Do it manually until it’s boring — boring means the decision has gone mechanical in your own head before it ever touches code. Write the rule down, not the vibe — if you can’t reduce the call to an explicit rule, you’re not ready to automate it, you’re ready to automate confusion. Automate the rule, not the outcome — the system should execute a decision you already know how to make, not discover a new one. Keep a human in the exception path — every process throws edge cases, and the businesses that get burned by AI are usually the ones that removed the person who used to catch the five percent that doesn’t fit the rule. Review the rule quarterly — markets move, and a rule built on last year’s judgment decays.
The leverage was never in the automation. It’s in the judgment that came before it. Software multiplies whatever you feed it — competence or carelessness, at the same speed. Most founders are in a hurry to remove themselves from the loop. I’m careful about which loop I remove myself from, and only after I’ve closed it by hand enough times to trust the rule more than I trust my attention that day.
That’s the whole edge. Not the tools. The sequencing.