I've Watched Founders Add AI to Broken Processes for Two Years. Here's the Pattern.

The businesses where AI changes everything share one thing. They knew what they were doing before AI touched it.

Before You Add AI, You Need to Answer This First

I facilitated two Work Smarter with AI workshops last week with ten business owners, all of them smart, all of them running real businesses, and every single one of them came in asking the same version of the same question. How do I use AI to speed things up?

Not one of them asked whether their business was ready for it.

That’s the problem I keep running into. Founders are buying into AI the same way they bought into creating a digital course five years ago, the same way they bought into a CRM before they had a sales process, and the same way they hired a team before they had anything for that team to run. They add the tool and wait for the chaos to clear. It doesn’t.

AI is not a clean-up crew. If your operations are held together by your personal attention, your memory, and your willingness to step in whenever something falls through a crack, adding an AI layer doesn’t fix that. It accelerates it. You get faster chaos, better-looking chaos, and chaos with a summary attached.

I’ve worked alongside founders for fifteen years, and the pattern is consistent enough that I am not surprised when I see it anymore. A $2M firm running on the founder’s instincts, a $5M firm where the team is waiting to be told what to do next, a founder who can’t take a week off without things slipping. These businesses aren’t broken in an obvious way. They run, they generate revenue, and clients are reasonably happy. But the founder is the operating system, and when you try to add AI on top of a fully human-dependent operating system, you don’t get efficiency. You get one more thing the founder has to manage.

The businesses where I’ve seen AI actually change the way work gets done share one thing in common. They already knew what they were doing before the AI touched it.

The unsexy work that makes everything else possible

My own team had a Monday morning problem. We’d meet at 11 am, assign tasks, talk through client work and upcoming milestones, and then the meeting would end. By Tuesday afternoon, maybe Wednesday, it became clear that half of what we discussed hadn’t moved. Not because anyone was lazy. Because the tasks never got created. They lived in the meeting, in someone’s head, and then they quietly dissolved before the next Monday when we’d meet again and go through the same conversation.

The fix looked simple from the outside. But before any automation was possible, I had to know exactly what the process was supposed to be. Who takes the notes. Where the tasks live. How they get assigned. What counts as an action item versus a discussion point. I had to have a system before I could automate one.

Once that was clear, I built a skill that runs every Monday at 12:15 pm. It pulls the Zoom transcript from our team call, reads the AI summary, identifies assigned tasks, and creates them in Notion with the right person attached. By the time I’m done with whatever comes after the meeting, my team’s tasks are already waiting for them. That process hasn’t slipped in weeks.

But if I had tried to build that automation before we figured out how we actually wanted task ownership to work, before we agreed on where tasks live and what the meeting structure should produce, the automation would have just automated the confusion.

That’s the pattern I see playing out everywhere. The founders who get the most from AI are the ones who’ve done the work of documenting what they actually do. The ones who are frustrated with AI are, almost always, the ones who skipped that step.

What the foundations actually look like

I’m not talking about a six-month operational overhaul before you’re allowed to touch AI. I’m talking about being able to answer three things clearly.

What is the process? Not the ideal version of it, the one that actually happens. Who does what, in what order, and where does it break?

What is the output supposed to be? If you can’t describe what done looks like, you can’t tell AI what to produce, and you can’t tell your team either.

Where does it live when it’s finished? A task that gets completed and doesn’t land somewhere useful isn’t finished. It’s just done in someone’s head.

Photo by Vitaly Gariev on Unsplash

Most founders can’t answer all three for more than a handful of their core processes. That’s not a character flaw, it’s just what happens when you build a business around your own ability to hold things together. You never needed to write it down because you were always there to fill the gap.

AI can’t fill the gap. It can follow instructions and pattern-match against what you’ve given it, but it cannot compensate for a process that was never defined. When people tell me AI isn’t working for them, nine times out of ten, the first issue isn’t the prompt. It’s that the underlying process was never clear enough to hand off to anyone, human or otherwise.

The work nobody wants to sell you

Nobody builds a course called “Document Your Processes.” There’s no waiting list for an SOP workshop. It’s unglamorous, it takes longer than expected, and it requires you to slow down before you can speed up.

But I’ve watched what happens on the other side of it. I’ve watched founders take a week off while their teams handled client work, not because they hired differently, but because the business finally knew what to do without them in the room. I’ve watched Monday meetings stop being the place where good intentions go to die, because the system, after the meeting, does the work that used to depend on someone’s good memory.

The boring infrastructure is what makes the interesting stuff possible. You don’t get the automation without the process. You don’t get the process without the clarity. And you don’t get the clarity by adding another tool. You get it by sitting down and answering the questions you’ve been too busy to answer.

AI is a powerful layer. But it’s only a layer, and whatever is underneath it determines what you actually get.

If you want to work through this hands-on, I’m running the next Work Smarter with AI workshop on June 18th at 12 pm ET. We’ll look at where your biggest operational friction is and build the AI layer that actually fits.