In the AI Era, the Question Was Never 'Can It Be Built' — It's 'Should It Be Built'
A friend in the import-export trade built a clean, fully-featured product in a week — two years ago that was two months of work for a small team. Third week after launch: single-digit daily active users.
AI has answered “can it be built” for everyone, but it never touched the two questions that actually decide life or death: does anyone want this; and even if they do, why would anyone ever see it.
The moment building became nearly free, those two questions went from being part of the business to being almost all of it.
Last week, a friend in the import-export trade showed me a product he’d just shipped: a quoting SaaS for small merchants. Design, frontend, backend, deployment — all of it him, alone, built in a week. Clean interface, complete feature set. Two years ago, this was two months of work for a small team. He was excited, and asked me whether it could make money.
Third week after launch: single-digit daily active users.
The hottest question this year is: with AI, can one person start a company? But that’s the wrong question.
AI has indeed driven the cost of design, development, and deployment to MVP down to nearly zero — you can spin up a polished product in a week. But think about it: it made “getting it built” free, not “building it well enough.” Whether you can build it is no longer the barrier. Whether you build it well enough still is.
But making the product good enough is only buying a ticket to the gate. What actually decides life or death isn’t how pretty the product is — it’s two other questions that aren’t even on the same axis as “how well it’s made.” And AI hasn’t solved either one:
First, does anyone actually want this. Second, even if they do, can you get it in front of those people.
A caveat up front: both questions assume the product is at least good enough. If it isn’t, neither question is worth raising.
The First Question: Does Anyone Want It
My friend’s product died right here. Even though he doesn’t see it that way.
CB Insights dissected several hundred startups that went under. Among the causes of death the founders wrote down themselves, the number one — at 42% — was “no market need.” On the surface, 70% of companies die because they ran out of money. But running out of money is the outcome, not the disease. The real disease, most of the time, is that nobody wanted the product in the first place.
A quoting tool that genuinely hits a pain point and found the right people doesn’t die in its third week just because nobody promoted it. It would have at least a small base of users who come back on their own, who pull in their colleagues on their own. Single-digit daily actives doesn’t mean he’s missing a salesperson — it means these people either don’t exist, or he hasn’t found the right ones. “The product isn’t bad” is the founder’s self-assessment, and self-assessment is the least trustworthy kind of evidence there is.
AI drove the cost of “building” to the floor, but it didn’t lower the cost of “figuring out what to build, for whom.” If anything, because building is now so fast and so satisfying, this step is even easier to skip than before: the product is already live before anyone asked whether there’s demand.
Here’s a positive example. In 2013, the Stanford folks who would go on to build DoorDash wanted to do something around small merchants — but they admitted that, at first, they “had no idea what these merchants’ problems actually were.” So what did they do? They asked, store by store. They visited two or three hundred shops around the Bay Area. Until a shop owner in Palo Alto dug out a thick stack of delivery orders nobody could fulfill for her, saying she opened her shop to make food, not to deliver it. That’s when it clicked: putting takeout online had been done for ages — Grubhub, Delivery.com, those platforms help you discover restaurants, see menus, place orders. But what they brought online were restaurants that already kept their own delivery crews. The real untouched hard problem was the small and mid-sized merchants who had no drivers of their own: they could make the food, but who picks it up, who delivers it, how do you dispatch during the rush, how do you keep it on time. And that suburban, small-merchant market everyone underestimated was exactly what DoorDash set its sights on.
By common sense, the next step should be to build a delivery platform, right? They didn’t. One Saturday, in about an hour, they slapped together a static page, paloaltodelivery.com, posted PDF menus from 8 restaurants, and instead of online ordering, orders went straight to a Google Voice number the founders shared. Delivery fee: $6, no minimum. Why launch? In their own words, just “to see if people cared” — to find out whether anyone actually cared, whether anyone would really place an order.
About 45 minutes after the page went up, the phone rang. A complete stranger, someone they’d never met, wanted a Pad Thai from a Thai restaurant. They called the restaurant to place the order on the customer’s behalf, drove Tony Xu’s Honda to pick it up, delivered it by hand, and collected payment with Square.
Look at that — a few Stanford computer science students, perfectly capable of building something far more respectable, deliberately made the product as crude as it could possibly be. Because what they were validating was never “can we build a delivery platform” (they could; that wasn’t a question for them), but “is there actually anyone in Palo Alto willing to pay for this kind of delivery.” A stack of PDFs and a phone number compressed demand validation, fulfillment experimentation, and user insight into one minimal closed loop. Even after they saw people ordering repeatedly, what they thought wasn’t “this counts as a company now,” but “let’s keep going.” During their Stanford stint they hadn’t even registered the company — they only formally incorporated after getting into YC.
The key point: back then there was no AI, and they still minimized “building” and threw all their energy at “does anyone want it.” Today AI has already answered “can it be built” for everyone — so there’s even less reason to skip this question.
The Second Question: Why Would Anyone See You
Even if you found the right demand and built the product, you still have to get the people who should use it to see it first, then use it. And this step is the most counterintuitive: AI has now made “building channels” nearly free too — but the result isn’t that everyone is easier to find, it’s that the channels got collectively trashed, and everyone is harder to find.
The logic isn’t complicated. Think about it: once the cost of writing content and sending outreach approaches zero, everyone floods the same set of channels at once. AI mass-produces SEO, the supply of content in search results explodes, and organic traffic as a path is collapsing. AI blasts emails, inboxes fill up, and cold-email reply rates drop. Product Hunt is wall-to-wall AI demos, each link a flash in the pan no one remembers. But the total pool of attention is fixed. Supply explodes, attention stays constant, and there’s only one outcome: customer acquisition cost goes up.
Take Clay. In two years it went from $1M ARR to $100M, a $3.1B valuation, with OpenAI and Anthropic among its customers. It took over the prospecting, email-writing, and follow-up that a room full of salespeople used to do by hand, and compressed the cost of a single outreach to a hundredth of what it was. But it’s a tool anyone can buy. And anything anyone can buy can never be your advantage — it just makes that channel more crowded.
So what AI actually does is: it devalues the manual labor of “building channels” to the floor, while at the same time elevating “the one thing it can’t give you” to an unprecedented position. What is that thing? Resources: trust, relationships, audience. Anyone can buy Clay, but no one can buy away your rapport with target customers, your personal credibility, the group of people you have who are willing to listen to you. The moment AI trashes every open channel, it precisely raises the value of the closed channels — the ones gated by trust, the ones it can’t squeeze into. An introduction from someone you know doesn’t get buried under mass spam.
Then what about the Notions and Figmas — the ones that “grew without a sales team”? That’s product, not channel. Notion grew revenue 130x in five years; 70% of Figma’s large deals came from users spreading it organically. They didn’t rely on some growth playbook anyone can learn — distribution was welded into the product itself: collaboration tools, the moment you use them, force you to pull others in. But that’s structural luck reserved for a few products. Most products have no such loop — including my friend’s standalone quoting tool. They precisely confirm the premise: anything that can grow on its own must first be a product good enough to be worth sharing.
For the majority of products that won’t spread themselves, resources are the only starting-line advantage you have to offer that AI can’t give. One-to-one relationships — your contacts, warm introductions — get you your first hundred customers, your credibility, your foot in the first door. Broadcast trust — a founder with a hundred thousand followers, a recognized brand — is the compounding version of it: post a message and a hundred thousand people hear it. The former pries open the cold start; the latter rolls the snowball.
And precisely because it’s a starting-line advantage, it splits by lane. High price point, enterprise, cross-border, services — one relationship is worth a lot of money, and resources are close to the lifeline. But it doesn’t scale down to long-tail self-serve: you can’t sell a $9-a-month tool to ten thousand small business owners one coffee at a time. My friend happened to be in the latter lane: a product that doesn’t spread itself, and no audience or relationships banked in advance. So launch meant silence.
So the right way to phrase the second question isn’t “which channel should I use,” it’s “do I hold any resource AI can’t give, enough to pry open the hardest cold start.”
One more thing: resources are the stock you’ve accumulated. But people with no stock aren’t completely out of the game either. There’s another path that costs nothing, call it riding the wave. The total pool of attention is fixed, but it gathers in bursts onto certain events, certain people — and you can attach yourself and siphon off a little.
I know people with no resources who made noise purely by riding waves. The most direct kind: time your launch to a big company’s release day, shipping a product with similar features the same day. That day the big company draws the whole industry’s gaze, and standing in the same spotlight, you get caught in a sliver of it. A step further: on the big company’s launch day, actively find your points in common with it, package up that connection, and ship them together — borrowing its heat to stick a label on yourself. And the lightest kind: reply under an industry heavyweight’s post, and as long as your reply has substance, a single response or follow from them channels their audience over to your side.
All three spent no money buying channels and banked no audience in advance. They do the same one thing: attach yourself the instant attention temporarily gathers, borrow someone else’s momentum, and hand yourself out. It isn’t reusable the way holding a hundred thousand followers is, but for a brand-new product with nothing, it’s one of the few paths still proven to work right now.
Back to That Friend
What he lacked wasn’t technology, and wasn’t the product. What he lacked were the two questions he should have thought through before typing the first line of code: does anyone actually want this; and even if they do, does he hold any resource AI can’t give, that could get it in front of those people.
With neither, no matter how polished the product, it’s just a link nobody knows about.
The moment building became nearly free, those two questions went from being part of the business to being almost all of it.