Anyone can build it now. Then what?
AI made building software fast and almost free. The code was never the hard part. Choosing, finishing and distributing is where the value was hiding all along.
You describe what you want and, a few minutes later, it exists. A dashboard, a database, a page with buttons that click. I’ve been building small things with AI agents for months now (tools for the family, experiments, ideas that used to stay in a notebook), and the feeling never gets old. You see the thing working before lunch. The dopamine is real.
That feeling is also why I keep thinking about a recent episode of The Atlantic’s podcast, The Most Interesting Thing in AI. Nicholas Thompson interviewed Paul Ford: programmer for 30 years, entrepreneur, and author of “What Is Code”, probably the most famous magazine piece ever written about programming (or so it’s recently told). Ford described the moment he understood his industry was changing. He was vibe coding on a plane, watching a system assemble itself “like watching a house get built in five minutes”, and walked off to tell his business partner: “Death is coming.”
Then he arrived somewhere more interesting. “The code was never truly the hard part. The hard part was pushing things forward, coming up with new and novel ideas, communicating with consumers, building really good products that people wanted to use all day.”
(Dax Raad, co-founder of OpenCode, usually it’s nuanced and serious on these topics and was spot on with this tweet)
Code used to be expensive, so we assumed it was the valuable part. Ford puts numbers on it. A custom system he would have charged $400,000 for five years ago, with a full team and a four-to-six-month timeline, is now a one-week prototype at roughly an eighth of the cost. When the expensive layer collapses, you see what was underneath it the whole time. From what I’ve been building, and watching others build, the value was hiding in three places.
Choosing
When software cost six months and a team, the price forced you to think before committing. Now any idea can become a prototype by lunchtime, and the prototype is convincing: the interface exists, the charts animate, it feels like progress. Sometimes it is progress. Sometimes it’s a very attractive distraction. I wrote in May that creating got easier and meaning got harder, and this is where it bites first. When you can work on anything, choosing what deserves your weeks becomes the first real act of judgment. The ability to make something proves nothing about whether it was worth making.
Making the actual thing
Ford runs a software company, and even he admits this part stayed slow. He spent two months and hundreds of dollars in tokens on one unusual product, talking with the bot every day. “The code can happen really, really quickly, but the actual making a thing is still really, really hard.” The back office is still the back office: does it work as intended, does anyone return after the first visit, what breaks when real users arrive. Answering that takes iteration, time, and more perspectives than your own.
That last part is the easiest to forget, because an agent makes you feel self-sufficient from day one. Ford has a great phrase for what happens when everyone suddenly gets everyone else’s skills: “a festival of vengeance”. Product managers ship apps, designers build platforms, engineers skip the design review. His conclusion lands on the opposite side: “I think we’re actually all going to have to work together again in order to build good product.” You can do almost everything alone now. You usually shouldn’t, and convincing yourself of that is part of the work.
Distribution, and knowing what you have
The third place is the one I’d underline twice. The Internet is filling with AI-built products, newsletters, videos and agents, and attention is not growing to match. Having something to show is no longer unusual; having someone care is.
My favorite story in the episode is about this. Around 2015, Thompson, then at The New Yorker, suggested building an editing tool from the magazine’s archive of drafts and edits. Grammarly, but distilled from the best editing culture in publishing. It was too expensive, and they passed. Today the code would be a week of prompting, and what made the idea valuable was never the software: “the expensive part is that database and that knowledge and that little moat of knowledge that the New Yorker has about how to edit”, as Ford put it.
Even a small personal project has a moat: the reason you built it, the thing you know that others don’t, the audience you understand. Distribution starts with being able to say that clearly. What is this, who is it for, why should anyone give it five minutes. AI can draft the posts and cut the videos, but the intent has to come from you. As I wrote in May: AI can help produce the work, but humans still need to provide the reason for the work to exist.
The slow part
None of this is an argument against the tools. I use them every day, and they keep getting better. But the honest summary of this moment is the one Ford got to: the demo now takes an hour, the product still takes months, and an audience still takes years. He spent three decades thinking code was his trade and found out the trade was judgment all along. I keep arriving at the same place through different doors: humans are still the point.
Some references
The Economist: Choosing where to focus is among the most important skills
Yann LeCun also believes this: Anthropic’s War on Opensource AI
I wrote about it in a meaning perspective before: Creating got easier. Meaning got harder