I don’t go home often. But when I did recently, I installed an AI chatbot on my dad’s phone. He nodded politely, and I didn’t think much of it. A few days later, I walked in on him fixing a washing machine.
My dad is the kind of guy who fixes everything himself—air conditioners, plumbing, light fixtures. His toolbox is better stocked than mine. But this time, the machine was throwing an E3 error and the spin cycle was dead. He tore it apart and couldn’t find the issue. So, he did the unthinkable: he pulled out his phone and asked the AI.
The AI spat out seven possibilities, ranging from a blown capacitor to a clogged drain pump. Then, at the end of the list, it asked a question back: What is the model number of your machine?
My dad put the phone down. He stared at the screen, turned to me, and asked, “Is this thing actually thinking?”
As an interaction designer, I didn’t know how to answer him. But on the drive back, his question kept echoing in my head. It was a better, more accurate summary of the AI revolution than any research paper I’ve ever read.
The AI revolution isn’t about machines getting smarter. It’s about machines learning how to guess.
For the last forty years, we have operated under a strict, unspoken contract with our technology: Command and Execute. You click “Save,” the file saves. You type “Apple,” the search engine gives you apples. It doesn’t give you pears because it’s having a bad day. Designers have spent decades perfecting this contract—where to place the button, how to draw the icon, how to deliver feedback. But the contract itself never changed: the user commands, the machine executes.
AI shattered that contract. You type a prompt, and it might return absolutely anything. You ask it a question today, and it gives you a brilliant, nuanced analysis. You ask it the exact same question tomorrow, and it might confidently invent a fake fact. And it doesn’t know the difference.
We didn’t build a flawless oracle; we built a brilliant gambler. And now we have to learn how to negotiate with it.
Our relationship with software used to be “Command and Execute.” Now, our relationship with AI is “Guess and Negotiate.”
When my dad asked if the AI was thinking, he was hitting on the exact tension of our era. The AI wasn’t thinking. It was guessing. Guessing what you want, guessing how to phrase it so you’ll be satisfied, guessing whether it should ask a follow-up question to narrow down the odds. Old software never guessed. It just obeyed.
This is the real, massive design challenge of the AI era. It’s not about learning new tools or memorizing prompt structures. It’s about relearning how to interact with something inherently uncertain. It makes mistakes. It won’t volunteer an apology. Sometimes it’s smarter than you, sometimes it’s dumber than a rock, and you can never entirely predict which one you’re going to get.
Trust in the deterministic era was earned by never failing. Trust in the AI era is earned by how gracefully we handle the inevitable failure.
We have to stop treating AI like a super-calculator and start treating it like an unpredictable partner. The interface isn’t just a dashboard anymore; it’s a negotiation table.
The washing machine eventually got fixed, by the way. My dad ignored the AI, tore the machine apart again, and found a worn-out plastic gear. But something had fundamentally shifted in him. The next time something broke in the house, he didn’t reach for his toolbox first. He reached for his phone.
He wasn’t looking for a tool anymore. He was looking for a partner.
FAQ
Q: If AI is just guessing, isn't it inherently unreliable for serious work?
A: Yes, if you treat it like a traditional database. AI is probabilistic, not deterministic. It's unreliable if you expect 100% execution, but highly valuable if you treat it as a brainstorming partner that needs human verification.
Q: What's the practical implication for designing AI products?
A: Stop designing interfaces that promise perfect execution. Build UX that anticipates errors, invites user feedback, and frames the interaction as a collaborative dialogue rather than a one-way command line.
Q: Is the 'command-execute' era of software completely dead?
A: Not entirely. Traditional software is still best for deterministic tasks (like saving a file). But for any task requiring ambiguity, creativity, or synthesis, the deterministic contract is dead. We are now in the 'negotiation' era.