I have kept a running list of ideas since I was about nineteen. A plain text file, then a Notion database, then a series of notebooks with half-finished sketches. The list has never gotten shorter. If anything, having more skills, design, code, writing, made it longer, because more things became imaginable. Every new capability revealed three new things I did not yet have the time or money to build.
I suspect this is a familiar experience for anyone who thinks creatively for a living. The bottleneck was never the idea. The bottleneck was everything that came after it.
Why I Learned to Build
Part of what drew me to design and engineering in the first place was a specific kind of leverage. A painter needs a canvas and a lifetime of practice. A novelist needs language and solitude. A product designer who can also write code can take something that exists only as a mental image and, within a few weeks, produce something a stranger could hold in their hand and use. That gap, between imagination and artifact, is the most interesting space I know.
Learning to code did not make me a better designer in the narrow sense. It made me a different kind of thinker. I started sketching in the browser. I started treating prototypes as thinking tools rather than presentation tools. The executable artifact became the language of the idea. When you can make the thing, you think about it differently.
The distance between imagination and artifact is shrinking. What matters now is what you do with the time these tools give back.
The Collapse of Execution Cost
AI tools are compressing that same distance, only faster and more dramatically than before. I have been using Claude Code heavily over the past several months, and the cost of going from concept to working prototype has dropped by roughly an order of magnitude. Not the cost in dollars, though that too, but the cost in friction, in the activation energy it takes to start.
That matters more than it sounds. Most ideas die not because they are bad but because the first draft is too expensive. The evening you would need to rough it out competes against everything else that wants that evening. When the first draft costs a fraction of what it used to, more ideas survive the threshold, and more of the ones that deserved to exist actually do.
I have shipped three side projects in the past six months that, in a prior period of my life, would have sat on the list for years. I am not smarter or more disciplined than I was. The scaffolding just shows up faster, and I can stay in the state where I am solving the actual problem without dropping out to fight with infrastructure, boilerplate, or documentation.
What the Tools Cannot Do
I want to be careful here, because a lot of the hype skips over this: faster execution does not solve for judgment. It makes judgment matter more.
When the cost of building drops, the number of things that get built goes up. That is already happening. The App Store is filling with AI-generated tools. Product Hunt is drowning. There is a lot of software in the world right now that exists because it was easy to create, not because anyone needed it. The filter that used to be execution difficulty, the natural selection pressure of 'is this worth six months of my life?', is weakening. Something else has to replace it.
That something is curation. Taste. The capacity to look at your own ideas with the same skepticism you would apply to a stranger's and ask: does this deserve to exist? Who is it actually for? What does it do that nothing else does? These questions always mattered. Now everything rests on them.
Speed makes the good faster. It also makes the mediocre abundant. Curation is now the job.
Capital Still Has Teeth
Most of the discourse waves past the limits, so I will name them. Building is cheaper. Running things is not, at least not yet. API costs for capable models add up, and so does server infrastructure for anything that gets traction. The tools that speed up the creative work sit on a cost structure that has not compressed at the same rate.
This creates a particular kind of risk for the solo builder or small team. You can get to validation faster than ever before. But if the thing you built actually works, if it finds an audience, generates load, requires reliability, the economics shift quickly. Bootstrapping a consumer product to meaningful scale still requires either revenue, runway, or both. The zero-to-prototype gap is nearly closed. The prototype-to-sustainable gap is not.
In practice that means building things that can earn revenue early, things that are cheap to run at scale, or things so specific and valuable that a small number of paying users covers the infrastructure. The old advice, find a real problem and charge for the solution, still holds. If anything it holds harder now, because competition for free and generic has never been fiercer.
Standing Out in a Blue Ocean That Is Rapidly Turning Red
There is a version of the near-term future with ten thousand nearly identical AI writing tools and ten thousand task managers with an AI sidebar, all built by capable people using the same models, infrastructure, and distribution channels. Most of them will be fine. None of them will matter.
The blue ocean strategy literature talks about creating uncontested market space: finding the dimension nobody else is competing on and owning it before the crowd arrives. When building gets cheaper, the crowd arrives faster and the window for an uncontested position gets shorter. Speed and specificity are worth more.
What still differentiates: real domain expertise, a working relationship with the people who have the problem, a distribution channel built before the product, and a point of view specific enough to repel most people and draw in the right ones. None of this is new. It mattered before AI, and the execution layer is no longer where you win or lose.
Pixels and Atoms
I have been thinking more lately about software versus physical things, pixels versus atoms, as people say. For twenty years, software ate the world partly because the marginal cost of reproduction was zero. You build the thing once and distribute it infinitely. Atoms do not work that way. A chair is expensive to make and expensive to ship.
AI is beginning to close that gap in the physical direction too. Not in manufacturing yet, not at the pace of software. But in design, in specification, in the translation from concept to fabrication file. The industrial design work I did in Copenhagen, the months of iteration, the careful hand-drawings, the back-and-forth with the manufacturer, is beginning to compress in the same way software development did. The tools that existed when I was at Bang and Olufsen bear almost no resemblance to what is available now.
I do not think this makes physical design less interesting. It makes it more accessible, so the interesting questions shift. The question was never really 'can you make this?' It was 'why does this need to exist, and for whom?' That is the whole game now.
The interesting question was never whether you could make the thing. It was always why it deserved to exist.
Three Timelines
The past: the bottleneck was skill. Learning to design, learning to code, learning to write, these took years and created natural filters. The people who shipped things were the people who had put in the time. This was not entirely good. It kept out a lot of people with valuable perspectives. But it created a kind of earned judgment. You knew what was hard because you had done it.
Now: the bottleneck is shifting. Skill still matters, I would argue it matters more than ever, because skilled judgment is what separates good work from plausible work. But the execution layer is no longer the barrier it was. The people who are building interesting things right now are the ones who can hold a clear vision, make fast and confident decisions under ambiguity, and stay close enough to real users to know when they are fooling themselves.
Coming future: the categories will get stranger, and more useful. The categories we use to organize creative work, designer, developer, writer, strategist, are going to blur and recombine in ways that are hard to predict. Most of the interesting people I know already work across all of them, and the boundaries are getting more inconvenient for everyone else.
Possible futures, plural: the optimistic one looks like a Cambrian explosion of software. Thousands of tools built for audiences of hundreds, not millions, hyper-specific, hyper-useful, sustainable at small scale. The pessimistic one looks like a slurry of undifferentiated AI output flooding every channel until the signal-to-noise ratio collapses and people retreat to trusted human voices. Both are plausible, and both might happen at once in different corners of the internet.
The List
My ideas list is still there. It is still growing. The difference now is that the distance between an item on that list and something real and working is shorter than it has ever been. That does not mean every idea deserves to make the crossing. Most of them still do not.
But the ones that deserve to exist, the ones that solve a real problem for a real person in a way nothing else does, have a better shot than they used to. For someone who has been keeping a list since they were nineteen, that feels significant. It does not feel overwhelming or anxious. It feels like a door that used to be heavy now opens more easily.
The interface layer is where AI products win or lose. The model is a commodity. The design is the product. And right now, most teams are treating it as an afterthought.