Letting an AI agent run the pricing

Most companies aren’t ready to let an AI agent touch anything that costs money. I let one run the pricing for a business I own, and set it up so every change can be undone, every test has to clear a break-even fixed before it starts, and the calls that matter still come to me.

The business is Lost Horizon, the luxury glamping retreat I designed in the Texas Hill Country, and its two stays used to get their prices from PriceLabs, the property-management system and Airbnb at once, each running discounts the others couldn’t see. When I handed pricing to an agent I gave it one standing instruction, to keep seeking higher income and to keep improving wherever it can, and the rest of the setup is rules about how it’s allowed to do that.

What it’s trying to raise

The number is total net dollars, meaning revenue minus the cleaning I pay for on every stay and the channel fees. A rate rise that costs a few nights can look good on every per-night measure while total net falls, so a test only wins when total net beats the same period last year at the same lead time, checked against the market, and the gain clears the break-even fixed before the test began.

Every change comes with its way back

A scheduled agent runs every morning and makes one reversible change per run. Two experiments run together only if what they change and what judges them don’t overlap. Every experiment carries a hypothesis, a break-even and a decision date, nothing is judged before thirty days, since a week of stays at two units is too little data to mean anything, and the exact steps to undo the change go into the experiment log at the moment it’s made. After each change the agent checks the price a guest would actually pay, on the booking engine and on Airbnb, rather than trusting that the setting saved.

One file carries the memory

No session remembers the one before it, so one file carries the whole program, from what’s running and when each test gets decided to a list of things nobody may quietly retry. Every run reads it before touching anything and rewrites it before finishing, so each run picks up exactly where the last one stopped. Every run also has to say how it ended, whether it started or advanced a test, judged and closed one, or explained why waiting is worth more than acting, and “nothing to report” doesn’t qualify.

I make the calls

The agent does reversible, in-scope work on its own once I’ve said go, and anything material comes to me first. When it audited every booking channel, it found that Expedia had been adding a discount of its own. It laid out my options and waited, and I chose to raise the Expedia markup. Some rules are absolute, starting with never emailing anyone, and on the Gmail connector it reads, the send, reply and forward tools are blocked in my settings as a second lock.

What carries over

It runs each morning against the live calendars, with the next tests queued in the order the evidence points. Two things carry straight over to any team that lets AI make changes, whether it prices rentals or builds software. Every change starts with a baseline, a break-even it has to clear and a date to judge it by, and the calls that matter stay with a person. If you run rentals or lodging as a business, see how my team and I build AI into yours.

Read next

How AI-written code gets proven before it counts

How I build with AI