Kodak, AI, and the Right Question
Kodak built the first digital camera yet asked the wrong question. AI will answer anything instantly, but the real skill is in wrestling with a problem long enough to know what to ask in the first place.
In 1975, a 24-year-old Kodak engineer named Steven Sasson carried an eight-pound prototype into a conference room. It could capture a 100-by-100-pixel black-and-white image and write it to a cassette tape in twenty-three seconds. The first self-contained digital camera.
Management didn’t dismiss it. They knew it could be big. But instead of asking, “How will people want to capture and share memories in the future?” they asked, “How can this help us sell more prints?” That single choice of question locked them onto the wrong track.
In the HBR article Kodak’s Downfall Wasn’t About Technology, the authors point out that Kodak bought Ofoto in 2001, years before Facebook existed. They argue Kodak was within reach of the future but failed to see it. If the company had leaned into its own tagline, “share memories, share life,” and built Ofoto into a true social platform, it might have led the way in what we now think of as social photo sharing. Instead, Kodak used Ofoto to drive printing. In 2012 they sold it to Shutterfly for less than $25 million. That same month, Facebook paid $1 billion for Instagram, a company built on the exact antithesis of printing photos.
Fast-forward. Kodak has a niche resurgence (and I'm loving it). Gen Z is buying film again — Kodak Gold, Portra — for the slower, more tangible process. Photographers are even trying to capture the wave of film by selling digital film packs to mimic the look of these film stocks.
At least for Kodak, this question was better: “Who still values film, and why?” It carved out a small but passionate and dedicated market. And yet, even with rising film sales, the company warned investors it may not make it. Losses, debt, a narrowing runway.
Same company. Same industry. Two completely different arcs, each shaped by the opening question, the wrong question.
Peter Drucker once said the most dangerous thing isn’t wrong answers — it’s wrong questions. And now we have a tool that will answer anything we ask, even the wrong ones. Instantly.
The real friction isn’t in debating answers. It’s in the slower, less glamorous work of shaping the question itself. You have to live with the problem long enough to understand what it is you’re really asking. That means digging into context. Knowing the history and what’s been tried, what’s failed, what’s quietly worked. Letting ideas clash until you feel the shape of the gap you’re trying to close. It’s awkward and often frustrating. AI will happily skip that step for you, delivering something that looks like progress. But without the grind that got you to the right question, the answer has no real weight.
IMD professor Arnaud Chevallier calls it your “question mix.” He outlines five kinds: investigative (“What’s known?”), speculative (“What if?”), productive (“Now what?”), interpretive (“So what?”), and subjective (“What’s unsaid?”). Most of us overuse one or two and neglect the rest. That’s fine until the problem you’re facing needs a different lens and you don’t have it.
When I work through an issue, I’ll run “what if” chains in my head: If we do this, it triggers that, which sets off something else… and where does that leave us? If I hand that to ChatGPT too early, it will send back something "perfect". But perfect isn’t the point. The point is whether that chain is even the right one to be thinking about. ChatGPT will just happily oblige and give you all the right (and sometimes terrible) answers to the wrong questions.
Cal Newport calls this “time under tension” which is the mental equivalent of holding a heavy weight for a few seconds longer than you want to. That’s where the strength comes from. Skip it and you’ve done the motion without the work.
That’s why I push AI later in the process. Frame the question first. Wrestle with it. Talk it over with people who will push back. Take a walk thinking through what’s actually at stake. Then feed it to the machine. Ask it to poke holes, find counterexamples, surface patterns from places you’d never think to look.
Kodak’s story is a cautionary one. They had the right tool at the wrong question in 1975. Found a better question decades later and still might lose. The right question doesn’t guarantee success, but the wrong one almost guarantees failure and wasted time and resources.
In a world of infinite answers, the rarest skill isn’t finding one. It’s knowing which question is worth asking.
