Take a routine care plan: plugin updates and small fixes across a dozen WooCommerce stores. The work used to fill an afternoon. It takes under an hour now, because a script does the sweep and a coding agent drafts most of the diffs I review. No client fired anyone. The invoice is just smaller, and that is what AI productivity gains look like from the shop side: the margin moves to whoever set the price.
Marco Baity-Jesi argues on Towards Data Science that AI does not decide who loses work, companies do. The real question, he says, is who captures the gains. His thought experiment is simple: one worker becomes ten times more productive, demand stays flat, and either nine colleagues go or ten people work fewer hours. The technology does not pick the outcome. Contracts, bargaining power, and institutions do.
Who captures the AI productivity gains
For a freelancer or a small shop, the institution that decides how the gains get split is the contract you signed. Bill hourly and the client captures the gain by default, because the task takes fewer hours and the invoice shrinks on its own. Bill for outcomes and you keep it. The productivity-pay gap research from EPI shows the same pattern across the whole economy: pay has lagged productivity for decades as the mechanisms that passed gains to workers weakened. In freelance work that mechanism is pricing, and you control it. (Less true when you sell to agencies that still buy in hours; procurement does not reprice fast.)
Don’t sell the same hours cheaper
I have seen two bad responses to this. One is dropping your rate as tasks shrink, which makes the work cheaper twice: fewer hours, at a lower price per hour. The other is billing the full hours anyway and hoping the client never asks why an afternoon of updates now takes less than an hour. If your income is hours times rate, and tools cut the hours, you have built the “nine colleagues go” outcome into your own business. Refusing the tools just means the next shop bids lower. The tools are not going away, and WordPress even has a dedicated AI team building an official client and plugin.
Sell outcomes, keep the tools
Care plans priced around outcomes set the right default. Scope what the client is buying: update windows, a test order through checkout after every WooCommerce release, plugin conflict triage, a backup restore test once a quarter. When agents cut your delivery time, your margin widens instead of your price falling. The freed hours go to judgment work nobody gets from a model: reading a changelog that adds tracking scripts, or deciding a plugin does not belong in the stack.
Where the tooling runs matters too. Agents belong in your pipeline and on your machine, not on client sites. API keys should never sit in the options table. I wrote about what AI coding agents should never run on client sites, and the same rule applies here.
If you want help turning a maintenance offer from hours into outcomes, or deciding where AI-assisted work should live in your pipeline, that’s the kind of engagement I take.
Start with one line item this week. After the next WooCommerce update on every retainer site, place a test order through checkout and confirm the order email arrives. That check is why clients keep paying a person, and it is a good first outcome to name in the scope.
