What Changes When Agents Can Check Out?
Bob Michaels explains how AI summaries change search visibility, how he keeps client work moving, and how AI agents can handle checkout and help build a store. The essays share a question: who gets to act, and what information do they need?
Written up in full
Transcript
You search for something your business sells. A summary appears at the top, gives you the answer, and you never visit the websites underneath it.
That's convenient when you're the person searching. It's a different feeling when one of those websites is yours.
Hi! I'm Bob Michaels, and this is Evolving the Web. Every two weeks I talk through the essays I just published, plainly, as one story. This time I've got four essays, all about how AI is changing who gets attention and who gets to take action.
The first starts with a real experiment. Researchers hid AI summaries from some people and showed them to others. When the summary appeared, clicks out to websites fell by about forty percent.
That changes what visibility means. The summary is doing some of the work your website used to do: answering the question before someone gets to your page. So being named and quoted in that summary can matter more than winning a place in the list below it.
How do you tell whether your company is showing up? I ask the main AI assistants the same real buying questions on a fixed schedule. Then I check that a mention really means your company, not another business with a similar name, and ask the model to say why it chose that answer.
That last part is useful. A name showing up is one thing. Knowing the reason gives you something you can inspect, instead of a result that feels like a fortune cookie.
Search is about whether people can find and trust your business. The next essay turns to a quieter problem: once work is underway, who has to do the next thing?
I run about a dozen client projects at once, with AI agents doing much of the work. HALO is the system I use to keep track, and it exists to answer one question: whose court is the ball in?
If I've handed the client everything and I'm waiting on them, that's a good day. The next move is clear. If something's sitting undone on my side, that's the failure the system exists to prevent.
A task can't just say "blocked" and leave the mystery hanging. It has to say who it's waiting on, me or the client. Before I call anything blocked, I ask what I could hand over to get it moving: a plan to approve, a draft to read, or one question with the options laid out.
That's a small change in wording with a practical effect. "Blocked" sounds like a closed door. "Waiting on you to approve this plan" gives the other person a handle.
The same question gets more serious when an AI agent can act on a checkout. At the end of September, Shopify gave browser-based AI agents three named checkout actions: look at the checkout, change it, and complete it, with the buyer's okay.
The agent doesn't have to guess which button to press or which box to fill in. It calls a clearly named step. That's a bit like giving someone a set of labeled controls instead of pointing them at a crowded dashboard and hoping for the best.
For any site that takes money or bookings, the question is plain: which actions should an agent be allowed to take, and which need a person?
The buyer's okay is part of the checkout actions Shopify described. Each business still has to decide where an agent can act and where a person needs to step in.
Now take the idea a step earlier. What if a conversation could create the store itself?
On October 1, Shopify launched Canvas. A merchant can build a store by talking with Shopify's AI assistant and see every page laid out at once. A Shopify director said someone can build a fully custom store in about twenty minutes.
That changes what website software is for. Its old reason was to let people who don't program make changes to their sites. If a conversation can build the site, that reason gets weaker.
But a conversation can't supply the facts behind the store. It can't know what you actually offer, what your prices are, or what proof you have. Someone still has to gather and check those details.
That's the thread running through all four essays. AI can answer a question before a click, keep work moving, take steps through checkout, and help build a store. The choices and facts still need care.
If you're curious, you can read the essays on bobmichaels.ai. I'm Bob Michaels. Thanks for hanging out. See you in two weeks.