I built more software for myself in the past two years than in the previous decade. Browser extensions, mobile apps that I use every day, a Trello clone, and hundreds of scripts. None of these would ever be products that I would have paid much for. They serve me, and that’s enough.

Making apps used to cost too much. When mobile phones were getting massive adoption, and I was starting my career as a software developer, people would often come up to me with ideas for apps they wanted to build. Their faces turned so quickly from excitement to disappointment when I replied “that would be at least a team of five, about 200k, and 4 months to get a prototype working”. Even simple ideas required serious investment.

That economic barrier is collapsing. LLM code generation fixes what I call the “Coding Army Conundrum”: to make useful, quality software, you used to need an army of developers. Thus, the only software that got built was software that could justify its existence economically.

My tools don’t have a business plan, but a clear purpose and detailed descriptions of how I want the system to behave. They might break, but an agent can fix them. And the cost of rebuilding is now low enough that maybe I won’t bother with fixing it, and just rewrite it.

Precisely describing what you want has always been a valuable skill. Now, it’s critically important to make the most out of this age of machine intelligence.