02: AI makes you feel good, then what?

Recently an acquaintance of mine dropped me a message. He wanted to design and build a site for a business his friend is running. It sounded simple on paper. A landing page, and a portal where customers could log in to book a slot and make payment. He asked for the timeline and quotation, got a sticker shock, and went MIA for the longest time.

Six months later, he came back.

The first thing I received was, “Hey, the design is mostly done, done by me on Claude. I just need your help with the dev and to put it online.” Earlier in the chat, he had also asked whether doing the design himself would make the project cheaper. The framing was clear. The expensive part, in his head, was done. What remained was the wiring, and the wiring was what he wanted me for.

I was already prepared to receive something non-Figma, because everyone can “design” nowadays right? I was expecting a full package. A repo, dependencies installed, the whole thing. Maybe a Loom walking me through the flows. Lo and behold, what arrived was a login page and a dashboard. That was it. Two screens. The two screens AI is best at making, because they exist on every site it has ever been trained on.

I sat with it for a while.

When someone tells me design is mostly done, I expect every flow walked through end to end. I expect a screen for when a payment fails. A screen for when a customer wants to cancel a session. A screen for password reset, for an empty state, for a teacher dropping out, for a refund. I expect the small, unglamorous decisions that decide whether a product survives its first real customer.

That’s the quiet thing about product work. The gold isn’t in the screens you see. It’s in the scenarios nobody asked about. The edge cases that aren’t fun to think through but eventually decide whether the product survives contact with a real customer. AI is fluent at the first part and silent on the second. It builds the version of your idea that looks like a product. It doesn’t build the version that behaves like one.

I clicked into the dashboard. It looked the part. Buttons, a table, a sidebar, the language of every dashboard he had ever seen. None of it worked. Nothing was wired to anything. And buried inside the file, AI had quietly suggested a subscription tool to handle the actual booking and payment logic. The screens were his. The thing that would have made them function was always going to be someone else’s software, paid monthly, already thought through by people who have run those scenarios a hundred times. He just didn’t see it that way. To him, the dashboard was the product. The SaaS underneath was a detail.

Deferring is not solving. The cost just moves.

AI made it easier than ever for him to start. The continuing is the hard part, and the continuing was nowhere in the file. $200 a month in tokens, a stack he can’t realistically maintain, and a SaaS quietly waiting to do the actual work underneath.

Go on Youtube or IG and you can find plenty

The thing is, the function he was looking for already exists. It’s available as a subscription for $50 a month, with every scenario already thought through by people who have run into them a hundred times. It was already inside his own file. He just didn’t see it that way.

He’s not naive. He’s human. This is what happens when a tool makes the easy sixty per cent feel like the whole job. He felt like a designer. He felt like a builder. He felt like he had a product. And he came back expecting it cheaper, because the design was done.

For six months, AI gave him exactly that.

Everyone is trying to build a startup now. Does everyone really need to?