Depth Is Not Demand

Where are the AI success stories?

I was talking with a good friend about what are the incremental gains promised by AI, specifically generative AI, over the last few years.

We wondered aloud, as consumers, where are the outsized benefits we were promised over the last few years? And at work, where has AI restructured our day so we can focus on more important strategic decisions, and less on the grunt work handled by the AI?

Outside of the big AI infrastructure providers, where are the huge AI success stories?

To wit:

• Is Google Maps 10% better at getting us to our destination faster / 15% more eco-friendly?
• Is Airbnb 20% better at matching our interests and improving the end-to-end trip experience?
• Is [X software / company] incrementally and/or an order of magnitude better at [y]?

What Has Changed?

Here's my answer to what AI has changed, if not "made better":

You know, the typical answers.**

There are a lot of opinions on the above. Instead, I want to spill some ink about what has changed for me, as a builder of software.

Speed Unlock

In 2026, the speed / ability to take your concept from "idea" to "pilot" to "production" has flattened roughly to the time it takes to deploy an agent to Vercel.**

AI can also, and I've been a big offender/victim of this trend, take me down wormholes where I'm convinced I'm building something really novel, only to realize I'm building a third-rate OPA (you needn't concern yourself with understanding what that means).****

So, AI helps me build fast, but it helps me build fast potentially useless things.

Nevertheless, I persist in building in anonymity a product that I think could help solve a real pain point around the messy middle between a tire kicker who wants more AI spend, a sales team who wants to provide more credit, and a CFO who wants to understand the bottom line of such pre-entitlement grants. (If that sounds like corporate gobbledygook, I guess it's because it is, so I need to work on my elevator pitch... Hold on, let me go ask AI...***

Depth is not Demand

More recently, I've been trying to step back from using AI to build a product of questionable market utility, to focus on using AI to research the problem and space itself, in order to better understand the category and ecosystem of the category pain points.

The results have been interesting -- apparently there is a lot more competition than I realized 2 months ago when I used AI to convince myself that the problem was worth solving / genuine.

One particularly useful quote from AI:

To me, "depth is not demand" means that just because I might know about a particular topic and be among the world's leading experts, and study something to the bleeding edge, this doesn't mean that there is any demand for the thing I'm studying outside of myself.

Also, AI is a fascinating confirmation bias machine that excels at convincing me that whatever train of thinking I'm going down is fantastically engaging, and that I'm among the leading experts on a particular topic, and that my analysis is at the bleeding edge ...

While a person might say, hey, stop talking about [x], I don't care... AI rarely will. And even if it does, you're only one prompt away from a whole new discussion ... So, it's a perfect tool to help you shape your pre-existing notion that the thing you're trying to solve is pretty important, that you're doing good work, and that if only I spent a few more days polishing this beta, then it'll be ready for people, and the people will love it ...

Important disclaimer:

It's never been easier to build something quickly that looks somewhat good, but is actually pretty useless. Not just to the people it's inflicted on, but to author itself, given the time spent building/prompting it into existence.

And, I think, this might be the answer to my friend and my question of "where is the great gain in software quality / functionality?": the features being built faster are not the features we want, much less need.

The current, in vogue terminology I'm circling around is the concept of "taste," which is, apparently, a hot commodity in short supply.

In other words, in 2026******, we have too many builders and not enough people with good taste about what's worth building.

Let's gloss over my defensible take where I decry tokenmaxxing as antithetical to building good taste, and skip to my thoughts on:

Problem-Market Fit

You didn't read that header wrong.

Stepping back, my definition of taste is the same as this one: choosing ideas worth building. With the cost of code racing to zero, the more important next step is to develop good taste to solve important problems.

Important aside, while the cost of the inference to generate the code that was previously written "by hand" (and/or copied from Stack Overflow) by human developers is not actually getting that much cheaper, rather, it's still being heavily subsidized during this phase. We're on the "customer acquisition" and "get them hooked" portion of this latest technology evolution.

The point remains -- it has never been faster to build a "reasonably functional" to-do list, or JavaScript calculator, or "control plane for agentic engineering" or "workforce management for a design start up." In a weekend or two, everyone feels like that they can be the next Linear / Atlassian / Loom, without thinking through the table stakes questions -- how does this serve the customers who would actually be interested in this?

Why wouldn't the company build your weekend project themselves and save themselves yet another recurring monthly SaaS expense?

And so, I don't think that "product market fit" (PMF) is the currency of 2026, it is rather identifying good "problem market fit" scenarios -- does your product meet a real enough problem that the company can't:

• buy it,
• build it themselves,
• extend what they already have.

So, in 2026, am I chasing a big/thorny/annoying enough problem that it won't succumb to what I call the Two Devs Claude Max Weekend test:

In my view, entire categories of software are going to be flattened / consolidated by the Great Inference Machines, but that's another post entirely ...

I am honestly not sure if Problem-Market Fit is just a (not so) clever turn of phrase, or if it is a useful question to ask. Perhaps finding product-market fit has always been about finding problem-market fit, in that it's paramount to identify a problem worth solving--"Make something people want," as Y Combinator might put it.

And so recently, in order to hopefully build something people want, naturally I've been turning to AI to tell me what people want. Why, you might ask? Because it's safer and protects my feelings!

And, like any other blog post / start-up advice / "realization" as anyone who has toyed with the idea of starting a start-up, I have come to the conclusion:

Nothing New Under the Sun

In 2026, I still need to talk to actual prospects, actual leads, actual people. Not AI. Nothing changes that. This essay does not change that. Talking to my friends to "validate" my product idea does not change that.

What I need to spend 10x more time doing is talking to the people who actually experience the problem, day-in-day-out. But how to find such people? (Perhaps that is what I should write about next ...)

If I'm honest with myself, it's not the AIs fault. It's always been, and always must be: buyer beware. If I believe the great Confirmation Bias machine telling me that I have built a "new calculus" even though I begged it "can you definitely, unequivocally, 100% confirm that you are not hallucinating!!! DO NOT HALLUCINATE!" ... That is not the tool's fault.*******

Although, it might be a cause for greater regulation -- either by governmental forces or the industry itself ...

And, if I'm doubly honest, the AI has warned me, this most common advice:

• Go talk to real customers,
• This chat cannot substitute talking to humans about their pain points,
• Go touch grass. (In so many words.)

But, I can ignore the AI, because it's just a machine. And it's easier to build something in private than actually pick up the phone or send an email and risk my feelings getting hurt. Or write an essay about it, and share it with a close friend, and face scrutiny ...

Conclusion

I am not sure what my call-to-action is. I have none.

I wanted to use my human brain to explore the topic of how AI is affecting our lives, and how AI is affecting my journey as a "builder" of a SaaS start-up company I'm not sure that anyone wants.

And to write something that is potentially interesting / useful to my friend, who inspired the conversation.

I could go deeper, but depth is not demand.

Footnotes

Note: apologies, I will use AI later to remind me how to turn these into proper footnotes:

*Is agent resourcing a category of resource planning, yet? If not, formally.
** I am assuming this is marketed as "almost instantaneous."
*** The wedge I'm chasing is: before card, pre-lead consuming real, costly AI compute for the company footing the bill). If you want to learn more, visit here.
**** But, if you're curious, Opa = Open Policy Agent, a leading "policy-as-code" framework for declaring / managing / enforcing your tech stack policies (official docs), though I prefer this definition as a kind of fancy allow-deny engine for policies. It's really an enterprise need only apply type problem/solution.
***** If taste sounds to you to be a kind of aesthetic choice a la Zen and the Art of Motorcycle Maintenance, then, I agree with you. ***** Don't worry, I'm among them, I can speak to it.
****** Gotta keep it fresh.
****** Thought it would stand to reason to provide some kind of regulation for AI providers to protect the interests of children / other folks who don't "know any better." Let's be honest, isn't it all of us?