An AI agent is three things: a goal, a set of tools, and a loop. This talk takes one apart on stage and shows that most of what is inside is code you have already written.
Two thirds of the time is the concepts, in plain terms. The model in the middle never does anything itself. It picks a function and fills in its arguments, and your code does the actual work. In WordPress that means your tools are the REST routes and action hooks you wrote years ago, and the model is a component you can swap out later.
The working example is a WooCommerce store, because it makes the cost obvious. Refunding one order takes six screens: Orders, filter by status, find it, scroll, Refund, line item, quantity, amount, confirm. Anyone who has run a store knows that path by heart. In the last third I replace it with a typed sentence, live: press Cmd-K, type “refund order 1042”, read a preview of what is about to happen, confirm. Then I show the code behind it, which is about a dozen ordinary PHP functions with a JSON schema for each. It is far less impressive than the demo, and that is the point. Nothing in it is specific to commerce; a membership plugin or an editorial workflow is put together the same way.
What you will take away: a working picture of what an agent is made of and how much of it you already own; why the model in the middle is a replaceable part, and what that means when you choose one; how much of the reliability comes from your code rather than from the model, the confirm step included; and an honest account of where this falls over today.
For developers who know WordPress and have not built an agent yet. No AI background needed.
One disclosure: the plugin I demo, Hoobert, is mine. It is free on the plugin directory with no paid tier, and the talk teaches the pattern rather than the plugin. The same code works without it.
// on_stage





