The interface looked polished. The feature list was long. There were workflows, dashboards, automations, and enough functionality to make the product feel more mature than it was. He told me he had built the whole thing with Claude over a weekend — and he wasn’t a software engineer.
He was excited, understandably. A few years ago, getting this far would have required engineers, money, and time. Now he had a working prototype before Monday.
But no one in retail had used it yet. And he had never worked in the industry himself. He was building for a user he did not really know, in a context he had not lived, based largely on what he imagined the problem to be.
That moment captured the promise and the problem of this stage of AI. More people can now move from idea to artifact. But artifacts are not the same as evidence of need.
Building Has Been Democratized
AI is democratizing software creation. A founder, designer, operator, marketer, or domain outsider can now create software that would once have required a technical team. That expands who gets to participate in the creation economy and lowers barriers, broadens participation, and speeds up experimentation.
While AI expands who can build, it does not automatically expand the judgment required to decide what should be built.
Of course, the underlying lesson is not new. Lean Startup told founders to test assumptions before scaling. IDEO helped popularize human-centered design. Steve Blank told entrepreneurs to get out of the building. For years, the discipline of innovation has warned against falling in love with the solution before understanding the problem.
But AI changes the stakes. The old danger was building the wrong thing after spending too much time and money. The new danger is building the wrong thing quickly, beautifully, and with enough synthetic evidence to feel right.
From Minimum Viable Product to Minimum Valuable Problem
If a founder can now describe an idea to a coding assistant and get back functioning software that looks convincing, it can obscure the fact that the underlying customer understanding is thin.
This shifts a central question for innovation. Increasingly, the issue is not whether a team can produce a minimum viable product. As Nick Coster and others comment, it’s whether they have found a minimum valuable problem: a customer need with enough urgency, context, and consequence to warrant a solution.
AI can help teams get closer to that answer. It can, for example, summarize customer reviews, analyze support tickets, scan forums, identify complaints, map competitors, synthesize interview transcripts, generate personas, and surface patterns in consumer behavior. Used well, AI can make discovery faster and broader.
Increasingly, AI can also simulate the consumer. Synthetic panels and AI-generated personas can test concepts, pricing, positioning, and potential responses. BCG has argued that these tools can make consumer insight faster and more scalable. Recent work from Ipsos and a Cornell-hosted arXiv study on synthetic purchase-intent modeling suggests a similar pattern: synthetic consumers can approximate aggregate purchase-intent patterns and support early concept screening, but they remain weaker as substitutes for real customer contact, especially when the question depends on individual context, workflow, trust, or lived experience.
That makes these tools useful for hypothesis generation, but not validation.
Leaders still need to know when the signal is shallow, biased, overfit to existing behavior, or culturally tone-deaf. A generated persona is not a person. A summary of complaints is not the same as watching someone struggle through a real workflow. A pattern in online reviews is not the same as understanding the moment when a customer feels frustration, hesitation, embarrassment, distrust, or urgency.
That judgment is what I think of as appropriateness. Appropriateness is not just whether a product solves a stated problem. It is whether the solution fits the customer’s reality. Does it match how they already behave? Does it ask them to change too much? Does it create new anxieties? Does it feel trustworthy? Does it arrive at the right moment? Does it respect the social, emotional, operational, economic, or cultural context in which the problem occurs?
In retail, for example, a workflow that looks efficient in a demo may fail on a store floor because associates are busy, systems are fragmented, managers are overloaded, and no one has time to adopt another dashboard. The same pattern shows up in other industries: a tool may be technically useful and still fail because it misunderstands trust, workflow, incentives, or timing. These considerations are the difference between software that exists and software that gets used.
Build Quickly, but Build from Contact
The implication is not that teams should return to bloated research cycles, endless strategy decks, or months of analysis before action. AI should make innovation faster. But speed should be used to learn, not simply to launch.
For example, AI can be used to scan the landscape, generate hypotheses, identify patterns, and simulate responses. Then use human judgment to test whether those patterns reflect a real problem in a real context. Talk to customers. Watch behavior. Understand constraints. Look for workarounds. Ask what people have already tried. Study what they ignore, not just what they request. Then build quickly — but build from contact.
The founder on the Zoom call had accomplished something remarkable. He had created functioning software without being a software developer. That suggests a future in which more people can move from idea to artifact, and where technical barriers no longer decide who gets to participate in innovation.
But it also showed the limit of the moment. He had built an app for retail before he had built understanding with retailers.
“Build it and they will come” was always an unreliable theory of innovation. AI makes it even more cautionary because it makes building feel deceptively easy and insight feel artificially complete. Today’s question is whether we have earned enough understanding to know why anyone would care.
Click here for more columns by Gail Martino; if you enjoy this content, please consider connecting with Gail Martino on LinkedIn.
