My business can't run without me. Where do I start?

If your business can't run without you, the usual fixes rarely hold. Working harder, hiring one more person or writing a binder of SOPs all leave the real problem in place: the judgment the work needs still lives only in your head. Start with one outcome. Find exactly where it stops when you're not there. Write down the decision you keep making. Give a system or a person the direction, access and guardrails to make it. Keep the final approval until it gets boring.

That is the short answer. The rest of this article is how I think about each step, and what I'd avoid.

Why does everything still come back to me?

Owners often describe it the same way. The team can do the work, but every exception comes back to you. A client question, a price, a "does this look right?", a decision nobody wants to own.

It's tempting to read that as a people problem. Usually it isn't. In my experience it's fragmentation. Tools that don't talk to each other. Decisions that live in one person's head. Work carried by hand from one place to the next. I said it on a recording once and I still think it's the clearest way to put it: "Friction is just fragmentation internally."

Your business probably doesn't need more productivity. It needs less friction. Adding more output on top of a fragmented business just makes the fragmentation louder.

Is the founder always the bottleneck?

No. Sometimes the real constraint is somewhere else: not enough of the right people seeing the offer, a confusing path to buy, a handoff that drops work, an offer people don't value enough. I wrote about that in The founder is not always the bottleneck. Check that first. It's possible to remove yourself from a process and still have a business that doesn't grow.

But when the work genuinely waits for you, the cause is almost always the same. The standard was never written anywhere except in you.

Where do I actually start?

1. Pick one outcome, not the whole business. A quote sent. A new client onboarded. An invoice paid. A weekly report delivered. One thing that matters now.
2. Trace where it stops when you're not there. Walk the outcome from start to finish and ask one question at every step: what still needs me? Each stop will be one of three things: a decision only you make, information only you have, or a permission only you can give.
3. Start with the part that drains you most. Not the most impressive part. The part you dread. That's where you get your energy back first, and energy is usually what's missing.
4. Write down the judgment, not just the steps. SOPs fail because they list steps, and the work never comes back because of a step. It comes back because of a judgment call. Write the rule behind the call: how you price, when you say no, what "done" actually looks like, how you answer the questions you get every week.
5. Give it direction, access, permissions and guardrails. Whether it's a person or an AI agent, treat it the same way. Tell it the goal. Give it the information and the access it needs. Decide what it may do on its own and where it has to stop and ask. The point is that you're only needed when it's actually required.
6. Keep the final approval, for now. Let the system do most of the work and keep the last look for yourself. When approving becomes boring because it's always right, you've earned the right to remove that step too.

Should I hire someone or use AI?

Either can work. Both fail the same way.

If you hire into an unwritten standard, the new person becomes one more thing that comes back to you. If you give AI an unwritten standard, it scales the confusion faster. AI isn't magic. It scales a recipe. It doesn't write one for you.

So the order matters more than the choice. Write the recipe first. Then decide whether a person, a tool or an agent should run it.

What should I avoid?

Adding things. AI is the age of removing things, not adding them. If your AI setup keeps adding to the company (more tools, more output, more things to check), you're probably using it wrong for the long term.

Automating a broken workflow. It makes the mess faster.

Skipping the part where you learn how the business actually runs. I think about AI like an autonomous car. Learn to drive first, so you can tell when it's off route. Then sit as the observer and step in only when it's necessary.

How will I know it's working?

Two signals.

The first is a number you can count. Pick one for the outcome you chose: how many times a week it waits for you, or how many hours you spend on it. Write today's number down before you change anything.

The second is harder to measure and matters more. On a recording about AI agents I put it this way: "the best signal to indicate whether or not you're using agents effectively, is peace of mind." If you feel lighter and the outcome still happens, it's working. If you feel like you're now managing the system as well as the work, something is still fragmented.

A useful test: step away for a few days. Don't treat it as a vacation you're hoping survives. Treat it as a diagnosis. Whatever breaks or waits for you is the next thing to fix.

What this looks like for me

I'm running my own media this way. Agents do most of the production and publishing. My job has narrowed: I own the goal and the metric, I track progress, and I keep the system honest and improving. It's a different kind of leadership than doing the work, and it has to be learned like anything else. Some of it still comes back to me. That's the next thing on the list.

If you can't tell which part of your business to start with, that's what the free Flow Audit is for. It's a short diagnostic that finds the one bottleneck keeping your business dependent on you, and it may tell you the fix is a simpler tool or no purchase at all.