When Don Scheibenreif first appeared on Invisible Machines, generative AI was cresting and the trough was a forecast. He’s back, three years later, retired from Gartner, with the third edition of When Machines Become Customers out in the world. The trough is now full of false promises, and board pressure to adopt AI meaningfully is high. Meanwhile, his neighbor may be using AI more effectively than many enterprises.
Bottom-up adoption, top-down denial
One term that highlights the problem at hand: token maxing. The Meta-era leaderboard energy of “who used the most” has curdled into something that is costly in the near and long term. Robb’s inversion is brutal if you look at tokens as a labor substitute: then aren’t they really bragging about labor costs exploding while revenue barely moves? Don concurs. Token maxing is a lazy metric, in the same family as “who sold the most this quarter.”
Likely due to ignorance and/or fear, the more powerful metrics are going largely unchecked: productivity, reinvented processes, talent deployed or removed or hired, overall impact of the investment. Without systemic change, Don reminds—echoing Age of Invisible Machines—none of it works.
The adoption story isn’t “enterprises aren’t adopting.” It’s bottom-up adoption with top-down denial—Colonel Klink’s “I see nothing” for those old enough to remember Hogan’s Heroes. Sanctioned or not, people are using the tools, and management teams are lagging behind. CEOs are rapt (80% name AI as the top transformative technology) and talk freely of business-model disruption. Too often, what they seek to operationalize is still a pile of demos.
Boards want more tech-savvy CEOs. Don’s description of savvy is using the tools, asking good questions, and letting teams across the org see leaders using the tech, not just mandating it. Robb calls it use case zero for an AI team: teaching the company to fish, not fishing for every department.
When machines become customers, trust becomes the product
On the topic of machine customers, Don shares a TED-like prompt that Chris Howard dropped in his lap a decade ago: “What if an IoT device were a customer?” Trust remains the bottleneck for agents with wallets, and Don points out that Google and Apple Maps took more than a decade to earn the trust of drivers.
Interest in shopping agents is high, but full use is still low. Retailers are already blocking outbound AI from their systems while building their own agents that leave the garden. Target’s “you’re on the hook if Gemini errs” announcement kills incentive. Amazon Buy for Me goes outside Amazon; eBay says no agents. Walled gardens can’t scale machine commerce, and Don’s standard is platforms where humans and machines meet in free, fair trade. Otherwise we will be in a world where you get a smaller market of locked printers that only buy OEM ink, so to speak.
The sharpest distinction may be incentive alignment. Alexa, Siri, Buy for Me—these are essentially cool demos that work for the companies that made them. Don is waiting for the product that says: rent or buy an agent you train and insure, that acts in your interest. Until then, assume the assist is not yours. Josh extends the implication: when that product arrives, enterprises that never moved the needle on inbound AI must restructure for customers who never visit the website, only negotiating through a gatekeeper that treats marketing as noise and wants data. Don’s brand answer from a Coke/Quaker marketer’s past: maybe two brands, a human brand and a machine brand built around service commitments and pricing transparency, side by side.
Reinvention starts where automation stops
Once again, service design arrives as the answer. Don links it to Gartner’s customer effort score—how easy did we make it?—and employee effort score. AI can see patterns humans miss and lift the minutiae safely, so long as a human is still signing off. Frontline managers remain the linchpin, with Coca-Cola Fountain research and a Qualtrics footwear study pointing to the same winning combination: product training plus managers who care. Walmart pulling self-checkout is a human-interaction story as much as a theft story. Moments of humanity are not soft. They are the part of the offer machines cannot replace.
Don closes where platform builders should listen: Gartner called machine customers early and ChatGPT made the idea legible. The next wave is implementation—figuring out which technology platform you should build or buy, and what ethics and safeguards to put in place. Knowing when to quit a failing AI initiative is a success pattern. Lazy thinking like token boards and “I don’t understand it, so I’ll oversimplify and hope” is expensive now. Define objectives first, then assemble the tools.
Don Scheibenreif is co-author of When Machines Become Customers and formerly led Autonomous Business research at Gartner. Josh Tyson and Robb Wilson host Invisible Machines. Listen to the full episode for the extended thread on hyper hype cycles, crisis engineering, and agents that work for you—or open the Selling to Machines ideation.