I've sat through a lot of AI demos this year. Founder pitches, internal product reviews, investor showcases, community sessions. And somewhere around the fortieth one, I noticed something that I haven't been able to stop thinking about.
They all look the same.
Not superficially — the logos are different, the verticals are different, the decks are different. But the underlying imagination is identical. The same use cases, the same user flows, the same moment where someone types a prompt and the AI produces something that looks impressive but solves a problem nobody was urgently paying to solve.
I've started calling this the imagination gap. And I think it's the most underdiagnosed problem in AI product development right now.
What the Imagination Gap Looks Like
The imagination gap is the distance between what AI can do and what founders are actually imagining with it.
Here's how it manifests. A founder discovers a new capability — say, multimodal reasoning, or long-context retrieval, or voice-to-action flows. They're genuinely excited. The technology is real. The capability is meaningful. And then they build… a chatbot. Or a summarisation tool. Or a "personalised" onboarding flow that's really just variable substitution with a nicer interface.
Why? Because those are the demos they've seen. Those are the use cases that got funded. Those are the product categories where there's a clear, comparable market size to put in a deck.
The imagination gap isn't about intelligence. It's about the gravitational pull of what's already been imagined.
Most founders aren't thinking about AI wrong because they're bad at thinking. They're thinking about AI wrong because they're surrounded by a reference class of AI products that all made the same conservatively creative choices — and those choices got normalised into what AI products are supposed to look like.
Why This Happens — The Three Root Causes
1. We confuse demo-ability with value
A chatbot is easy to demo. You type something, you get something. The feedback loop is immediate, visible, and impressive to someone who hasn't seen it before. A summarisation tool produces a shorter version of a long document — the delta is obvious in thirty seconds.
The problem is that demo-ability and user value are not the same thing. The most valuable things AI can do are often the hardest to demo — because they intervene earlier in a workflow, change a process rather than automate a task, or produce outcomes that only become clear over weeks of use.
When founders build for demo-ability, they end up with impressive products that don't get used after the first three sessions.
2. We anchor on what AI replaced, not what it enables
A huge proportion of AI products are fundamentally replacements. Replace the human analyst with an AI analyst. Replace the human customer support agent with an AI agent. Replace the human content writer with an AI writer.
Replacement is real value. Don't dismiss it. But it's also the most contested space — because every incumbent SaaS company, every services firm, every team manager is trying to build the same replacement. The differentiation surface is tiny.
The more interesting question is: what becomes possible that was previously impossible? Not faster. Not cheaper. Genuinely impossible without AI.
3. We mistake user familiarity for product-market fit
Users are familiar with chatbots. They're familiar with document summarisation. They know how to interact with these interfaces because they've used ChatGPT, they've used Claude, they've used Gemini. There's zero learning curve.
But familiarity is not fit. Familiarity is the absence of friction in learning how to use something. Product-market fit is the presence of urgency in wanting it to exist.
The highest-fit AI products I've seen are the ones where users feel a small sense of loss when the product is taken away — not because the interface is familiar, but because the outcome it delivers has become load-bearing in how they work. That's a very different thing from "I know how to use this."
What the Gap Costs You
The imagination gap isn't just a strategic problem. It has a direct operational cost that most founders don't see until it's expensive.
When your AI product is doing what every other AI product is doing, your differentiation has to come from somewhere else — usually speed, price, or distribution. Speed is temporary. Price erodes margin. Distribution requires either a brand you don't have yet or a channel you don't own.
Meanwhile, foundation model providers — OpenAI, Anthropic, Google — are shipping new capabilities every few weeks. Each release makes some portion of your product obsolete. If your entire moat is "we summarise documents better" and the next Claude release summarises documents significantly better out of the box, you don't have a product anymore. You have a demo.
The founders I'm most worried about are the ones who have raised on the strength of a capability demonstration rather than a deep understanding of a problem. They're nine months from a model update that makes their demo irrelevant.
Closing the Gap — What I've Seen Work
I'm not going to pretend there's a formula. But there are patterns I've observed in the AI products that seem to have found something real.
They start with a workflow, not a capability
The most grounded AI founders I've worked with start from a deeply specific workflow — not "procurement" but "the three hours a junior buyer spends reconciling PO discrepancies against invoice line items every Monday morning." They know that workflow intimately. They've watched someone do it. And then they ask: where in this workflow is AI the most precise tool?
Starting from capability — "we have RAG, what can we build?" — produces generic products. Starting from workflow produces specific ones.
They take their unfair advantage seriously
Every founder has some combination of domain access, data, relationships, or distribution that a well-funded competitor can't easily replicate. The best AI products are the ones where the AI capability is inseparable from that unfair advantage.
If you have five years of experience as a sales manager and 50,000 real sales calls, and you're building an AI coaching tool for sales managers — the AI is not your moat. The 50,000 calls are. The AI is the distribution mechanism for your real advantage. Build as if the AI will become a commodity (it will) and ask what remains.
They obsess over the moment of irreplaceability
There's a moment in every good AI product where the user experiences something they couldn't have experienced any other way. Not "this saved me time" — "I couldn't have known this without this product." Find that moment. Make it happen earlier. Make it happen more reliably. Make that moment the product.
What This Means for AI Founders Right Now
I've been running product discovery sessions with AI-native founders through AIBoomi — 12 to 14 founders at a time, half a day, structured discovery. And the single most common thing I see is not bad technology. It's founders who haven't spent enough time with the problem to find the version of their idea that only they can build.
The imagination gap closes not through more exposure to AI capabilities, but through deeper immersion in a specific human problem. The question isn't "what can AI do?" It's "what does this person need to know, decide, or do — and what's standing between them and that?"
Answer that question with specificity, and the right use of AI becomes obvious. Skip it, and you end up with another demo that looks like all the others.
The technology is genuinely extraordinary. The imagination hasn't caught up yet. That's both the problem and the opportunity.