I sat down with Tom McFadyen at Retail Club, out in the sun with barely a cloud overhead. Tom runs McFadyen Digital, which has built marketplaces and commerce platforms for decades. His firm just released AI Best Practices for Commerce, its fourth book. Tom says the marketplace book they put out about five years ago was a bestseller.

We only had about five minutes, and three points from it have stayed with me.

The print edition is enormous. Tom puts it at about 200K words in textbook format with small type, running past 500 pages. It launched with 125 AI use cases. A year after they started writing, the online version tracks more than 500.

That gap is why McFadyen moved the content online and made it free at ai-best-practices.com. Tom was open that part of the aim is to establish the firm as a thought leader in a new space, and giving the material away is a reasonable way to do that.

I'd treat the 500 figure carefully. A use case count tells you how many ideas someone has cataloged. How many retailers have one running in production, or whether it pays back, is a separate question the count doesn't answer.

Agentic commerce gets the headlines, and the picture most people carry is an agent completing a purchase. Tom pushed back on that. Retailers have used AI for recommendations and fulfillment optimization for a long time. His team splits commerce into four phases they call market, sell, fulfill and support, and counted around 60 use cases in market and another 60 in sell.

The newer agentic work he described turns up in catalog enhancement and product planning at least as often as in the transaction. For a retailer deciding where to spend, the back end may be where agents prove their value first.

Tom said this example came up earlier at the event, and he applied it to himself. He has bought the same socks on Amazon for decades and is happy to let an agent reorder them. Ask his wife whether she'd let an agent pick her dress for a wedding the following weekend and the answer is no.

Replenishment is easy to delegate because the shopper made the real choice years ago and the agent just repeats it. A wedding outfit depends on taste and on what other people at the event will think, and that's much harder to hand to software.

Most retail sits somewhere between those two. Groceries and beauty will draw the line in different places, and the companies selling agentic checkout haven't shown where it falls for the bulk of what people buy. I'm not convinced they know yet.