Burn Tokens, Not Headcount: Fix the Workflow Before You Hire
I break down the phrase "burn tokens, not headcount" the way a small business should hear it. It means you redesign the broken workflow before you add another person, and you use AI for the glue work so your best people get their week back.
There's a phrase from a recent AI talk that small business owners keep hearing the wrong way: "burn tokens, not headcount." Heard cold, it sounds like the point of AI is to get rid of people. The useful version is more practical. Before you add another person to handle repetitive work, ask whether the work itself should be redesigned. People matter too much to waste their time acting like glue between systems.
In this solo episode, I dug into that one phrase because it gets twisted so often. The speaker who used it claimed some startups were reaching demo day with roughly five times more revenue per employee than companies did 18 months earlier. I wouldn't build a whole operating plan around one number from a talk, and there are plenty of failed cases too. But the direction is real. Small teams are getting more leverage, and AI is changing what a small team can produce.
Here's the trap I see all the time. A lot of small businesses don't have a people problem. They have a workflow problem, and they keep trying to fix it with people. The process gets messy, so they hire a coordinator. The handoff breaks, so they hire someone above that. Reporting takes too long, so they assign a person to gather it. Now you've got a pile of people chasing information around when the real fix was cleaning up the workflow. Sometimes hiring is an expensive patch on a broken process.
So the rule I keep coming back to is simple. Fix the workflow first, whether you're adding AI or adding humans. If the workflow is broken before the hire, the new person becomes part of the problem. They learn the messy process, or they invent their own shortcuts, and now there's another node of confusion in the system. The same goes for AI. Get the process clear, then decide what computation can handle and what a person should own.
What does "burn tokens" mean in practice? You use computation before you spend another human hour. The AI drafts so a person doesn't write from scratch. Automation routes the message so nobody has to forward it by hand. I got an email from a client recently where we'd built some skills into their flow months back, and they said it made their drafting so much smoother than dedicating 10 hours to it. They're still writing the work themselves. The AI pulls everything together so they can write better and get it out the door. That's the opposite of AI slop. Slop happens when someone tells the AI "write this for me" and sends it without checking. Give the AI your brand voice, your knowledge base, and the experience of your best people, and it starts drafting things that sound like your company.
Think about an operations manager at a service business. This person is smart, knows the clients, and can solve problems. But look at their week. They're checking intake forms, chasing missing documents, updating the CRM, sending reminders, rewriting status updates, and asking the same three people for the same detail every single Friday. By the end of the week they're exhausted, and the work wasn't even hard. It was glue. An intake form can trigger a summary, flag the missing info, draft a kickoff checklist, and surface exceptions instead of making that manager inspect every case. Now they have time to handle the exceptions, improve onboarding, and sit down with customers. You get time back for the stuff you always wanted to do and never could.
The guardrail is what keeps this humane. Automate the drag and leave the responsibility with a person. AI can draft the email while the human owns the relationship. AI can flag the risk while the human makes the call and signs off, because the human is the one who's liable. Start with work that's repetitive, visible, and reviewable. That's the sweet spot. Don't start with "AI approves refunds over $10,000" or anything near firing people or legal judgments. Pick the spot where everyone agrees there's pain, you can see the result, and a mistake won't dig a deep hole.
At the end of the day, this isn't fewer humans doing the same broken work. It's better workflows that let humans do better work. Most of these fixes take a week or two if you stay focused on small tasks. If you want a starting point, the AI Automation Opportunity Map on the site walks you through questions like the ones I covered here, free, and shows you what you might be leaving on the table. Figure out one next step that's easy enough to start tomorrow, and you'll know whether you should hire today or fix the workflow first.