Walk into ten growth-minded businesses right now and ask the owner what they’re doing with AI. You’ll hear some version of the same answer ten times.
“I use ChatGPT to help with emails.”
“My team has been playing with it for marketing copy.”
“I asked it to summarize a contract last week.”
“We’re figuring it out.”
That’s not AI in your business. That’s a search engine with better grammar.
There’s nothing wrong with using ChatGPT to help write an email. But it isn’t AI installation. It isn’t operations infrastructure. It isn’t going to compound. It isn’t going to produce a line on the P&L next quarter.
It’s a tool you’re using occasionally — the same way you used Google a decade ago — and treating the occasional use as if it were the whole strategy.
What ‘Using AI’ Actually Looks Like
Here’s the pattern I see across engagements. The owner buys ChatGPT Plus, maybe Claude, maybe Copilot. The team learns to write decent prompts. Six months later, someone asks what the business has gotten out of it.
In almost every case, the answer is the same. Nothing measurable. Some saved time on individual tasks. A few good marketing drafts. No installed workflow. No eliminated cost. No compounding output.
The owner — usually a smart, growth-minded operator — looks at the spend and the lack of return and concludes one of two things. Either AI is hype, or they’re bad at AI.
Both conclusions are wrong.
AI isn’t hype. They aren’t bad at AI. They’re using AI when they should be running on AI. And those are completely different categories of work.
Using AI looks like ChatGPT subscriptions, scattered prompts, individual time savings on individual tasks. It’s what happens when a tool is available but nobody’s installed it. There’s no operator. There’s no workflow. There’s no measurement. The tool sits there waiting to be invoked.
What ‘Running on AI’ Actually Looks Like
Running on AI looks like installed workflows that produce specific outputs without anyone invoking them. The customer intake form that auto-routes, auto-prioritizes, auto-drafts a response, and surfaces it for review. The weekly report that builds itself. The proposal that gets generated from a phone call transcript. The compliance check that runs against every contract before it goes out. None of these are tools. They’re operations infrastructure with AI inside them.
The first category produces saved minutes. The second category produces compounding leverage.
The first category requires a human to remember to use it. The second category requires a human to monitor it.
The first category is invisible on the P&L. The second category eventually changes how the business runs.
Most owners are in category one, paying for tools, and quietly wondering why the math isn’t working. The math isn’t working because the math doesn’t work in category one. There’s nothing wrong with category one — but it isn’t an installation. It’s a subscription.
The Shift Is an Operations Decision, Not a Tool Decision
The shift from using AI to running on AI is not a tool decision. It’s an operations decision. It requires sitting down and naming the workflows that should run without human invocation. The handoffs that shouldn’t require a Slack message. The reports that shouldn’t require someone to build them. The decisions that should be auto-surfaced with data attached.
That work isn’t IT work. It’s the work of an operator who is willing to redesign how their business actually runs.
Most owners haven’t done that work yet. They’re still in the using-AI phase. Which means most of the value AI can produce is still sitting on the table, waiting for someone to claim it.
The owners moving into the running-on-AI phase right now are the ones who will compound. The owners staying in the using-AI phase will spend the next two years paying for tools that never quite produce.
The era of using AI is ending. The era of running on AI is here.
Most owners haven’t noticed yet.
If you want to talk through where your business sits — using vs. running on — I’m doing a few of these conversations this month. Schedule here.