EXPERT OPINION BY BILL MURPHY JR., FOUNDER OF UNDERSTANDABLY AND CONTRIBUTING EDITOR, INC. @BILLMURPHYJR

Apr 19, 2026

Photo: Katy Blackwood/NurPhoto via Getty Images

Starbucks just rolled out a ChatGPT app that lets customers order coffee using artificial intelligence.

If it catches on, customers would open ChatGPT, tag @Starbucks, describe what they’re in the mood for — or upload photos of their outfits, or the weather outside — and the AI would suggest a few drinks they might like. Then they’d checkout in the Starbucks app.

“Customers aren’t always starting with a menu,” Paul Riedel, Starbucks’s senior vice president of digital and loyalty, wrote in the company’s announcement. “They’re starting with a feeling.”

Personally, I’m starting with a feeling of skepticism.

For one thing, in the roughly 60 seconds I devoted to this whole idea in service of this story, I couldn’t get the thing to work. For another, online reaction to the notion, at least as far as I’ve seen, has been mainly of the mocking variety.

I’ll save you the search on X, but basically: If people can’t figure out what drink to order at Starbucks without using AI, God help us all.

That said, maybe there’s something more going on here. Because old soul that I am, when I first heard about this kind of offbeat experiment, I immediately thought of another one from 32 years ago.

People thought that one was half-baked too — and as you’ll see: look where we are today.

Anybody remember PizzaNet?

In 1994, Jeff Bezos was working on Wall Street, Sergey Brin and Larry Page had no idea what to do with their lives, and Mark Zuckerberg was in 6th grade. But the folks at Pizza Hut saw the future — and they rolled out what history now records as the first physical product available for purchase on the internet.

The result was PizzaNet, a webpage where you could order a pizza in certain areas of California, using what today looks like the world’s most rudimentary web form, but at the time seemed like something completely off the wall.

The experience was not smooth. You filled out a web form with your name, address, and phone number. Pizza Hut called you back to confirm the order. A driver came and you paid cash at the door.

The LA Times called it “clever but only half-baked,” noting that the whole process wasn’t actually easier than calling your local Pizza Hut directly.

Pizza Hut has kept the archived page live. You can see it here: a black-and-white form asking for your name and street address.

It is also, in its way, one of the most consequential pages in the history of American commerce.

I don’t think Pizza Hut made very much money off this at the time, but we can look back now and see that there was something maybe even more valuable long term — namely, the lessons that came with being a pioneer in online pizza-purchasing — paying off now in the form of a $150 billion online food delivery industry.

Back to Starbucks

The minute or so I spent trying to get the Starbucks ChatGPT app to work are about all I’m going to devote to it. I don’t even go to Starbucks very often.

But I’ve been playing around with AI tools long enough to have a feel for how this kind of learning actually works. I’ve built simple apps, and used ChatGPT to plan ski trips down to the dime.

I’ve spent what might look like an unreasonable amount of time using AI to game out my World Cup ticket-buying strategy.

But for all the time I’ve spent, I’ve comforted myself thinking that these small aimless experiments might sort of be paying off in longer term experience and knowledge. One day I’m designing custom stickers; the next day I’m layering an AI interface over Google Sheets and developing an entire customer relationship management system to run my small business — for the cost of about 5 or 6 hours prompting and revising and $9 a month in hosting.

That’s how you actually learn a new technology. Not by reading about it — who does that? But instead by using it for things that feel slightly absurd, failing in instructive ways, and building from there.

I’ve spent a lot of time here on Inc.com arguing that one of the smartest things small businesses can do is to study much bigger companies, and steal their best ideas. But here, I think it’s almost the opposite advice: Big companies like Starbucks can serve themselves well by copying the things that smaller organizations are doing almost every day.

Small to big

Small businesses run these experiments out of necessity, not strategy. When you’re operating with limited resources and no bureaucracy to clear, you try things because you have to — because waiting around for certainty is not an option. That scrappiness turns out to be an advantage in moments like this one, when nobody actually knows what works yet, and the only way to find out is to start.

And even not-so-small companies are getting the message. Walmart, Etsy, Little Caesars, and Booking.com have also been building similar integrations into ChatGPT. The category has a name now: “agentic commerce” — the idea that AI tools will increasingly help people not just search but decide, and eventually transact, inside a single conversation.

Starbucks’s ChatGPT integration probably isn’t going to put a dent in its app or the drive-through. But it could provide chances to learn things — about how customers describe what they want in natural language, about where the handoff between AI and checkout creates friction, and about which menu items surface when people describe a feeling instead of scrolling a list.

That’s data you cannot get any other way except by running the experiment. I think that’s what Pizza Hut learned by launching PizzaNet back during the Clinton administration.

Somebody believed the internet was going to matter for commerce, and they wanted to understand what online ordering felt like before they actually needed to know.

Whatever you think of ordering coffee by describing your outfit to a chatbot, that’s probably the right instinct to have right now about AI: pick something slightly absurd, run the experiment, and see what you learn.