THE NUMBER
There are three ways to price any task you are on the fence about, and almost nobody runs the third one. An online guitar school cut its team from 48 people to 30, kept revenue flat, and saves about $250,000 a year. It moved sales and operations work to AI agents the founder built himself, and his verdict on the agents versus the people they replaced was “slightly better results.” Cheaper and slightly better is the whole case for automating everything, and it is exactly the instinct the rest of this issue complicates.
3 THINGS HAPPENING RIGHT NOW
Same tool, opposite decision: more AI, not fewer people
A short-term-rental management platform raised its AI spend about 50% and laid off no one. The founder says that without AI he would have had to triple his 65-person support team, so the extra spend cancelled roughly 130 hires he never had to make. His line: “it creates more work because the productivity is so much higher.” One operator priced this kind of work and cut people; this one priced it and grew the work instead.
A Yale economist’s rule for what to automate: does it clear its own cost?
Pascual Restrepo, a Yale economist, argues in a new paper for a national economics research body that the deciding factor for handing a task to AI is not whether AI can do it. It is whether automating it clears the task’s own cost. His model sorts work into two buckets: “bottleneck” work worth automating (energy, infrastructure, science) and “supplementary” work that stays with people. He puts customer support, hospitality and creative work on the stays-human side, not because AI can’t do it, but because paying to automate it does not pencil out.
THE DEEP DIVE
He priced the job two ways. There was a third way he never needed.
Spencer Handley ran an online guitar school with 48 people handling sales, onboarding and operations. When a new version of Claude landed, he rebuilt the software he ran the business on himself and moved much of that work to AI agents, with the remaining team supervising them. The team dropped to 30. Revenue held. He saved about $250,000 a year and dropped four paid tools in the process.
Here is the part worth sitting with. Handley priced the same jobs two ways: an agent doing them, or a person doing them. The agent came out ahead on cost and, by his own account, slightly ahead on quality. Most owners only ever run that comparison. Agent, or the way I do it now.
There is a third way to price any task, and he did not need to reach for it: not doing the task at all. What does it cost you if you simply skip it? Almost nobody fills in that column. When you do, some tasks you were about to automate turn out not to be worth doing by anyone. That is the option for any owner weighing a tool against a hire.
ONE THING TO TRY THIS WEEK
Week 3 of 4: Price it three ways
You have three ways to handle any task you are on the fence about: hand it to an agent, pay a person to do it, or not do it at all. Most people price the first two and never the third. Here is how to price all three on one real task, and let Claude do the arithmetic.
Pick one task you keep going back and forth on automating. On paper or in a blank note, draw three columns and label them: Agent, Person, Skip it.
Fill in what you already know, in plain words. Under Agent, the tool or setup you would use. Under Person, what it would cost to pay someone to do this, a rough hourly or freelance rate times the hours, not an existing employee but a hire. Under Skip it, write your own rough guess first: what do you think it costs you to just not do this task at all? Lost revenue, missed follow-ups, risk. Put a number on it, even a rough one, before Claude ever sees the sheet.
Open Claude Code and hand it the sheet. On a Mac open Terminal, on Windows open PowerShell, type
claudeand press enter. Then paste something like: “Here is one task and three ways I could handle it: an agent, paying a person, or not doing it at all. Cost out each column in dollars per month. For the ‘skip it’ column, tell me what actually happens if I stop doing this task and what that costs me, whether that is lost revenue, missed follow-ups, or risk. Then tell me which of the three is cheapest once everything is counted.”Read all three columns side by side and decide. Sometimes the agent wins. Sometimes paying a person wins. And sometimes the skip-it column is the cheapest thing on the page, and the honest answer is to stop doing the task at all. Any of the three is a good outcome. The only mistake is never having priced the third one.
Keep the sheet. Next time you are tempted to buy a tool, add the skip-it column before you pay. Stuck? Reply to this email. I’ll help.
WHAT’S COMING
Next issue: the one automation you would happily repurchase at five times the price, and the one you would not buy again at any price. Willingness to pay is the only honest measure of what a tool is worth to you.
Manu