Software for One

That framing landed with a lot of people. The strongest pattern was not "AI will replace software teams". It was developers and semi-technical users rattling off things they had finally built for themselves after years of letting ideas die. Calorie trackers, custom podcast players, protocol test tools, health dashboards, note apps, Chrome extensions, language-learning tools, CRM replacements, and a pile of tiny home utilities all showed up as examples of software that would never justify a normal development cycle but is suddenly worth doing. Several people said the real benefit was not writing code faster. It was getting past the activation energy, the bug-fixing slog, and the modern web and mobile stack overhead that used to kill these projects. The thread also put useful bounds on where this works. The winning use cases were low-stakes, narrow-scope, single-user or single-family tools where "good enough" is enough and the feedback loop is immediate. That lower quality bar is exactly why LLM code is viable here. Nobody needs a calorie tracker for one person to handle every edge case on earth or support a team of maintainers five years from now. A few commenters pushed the idea further into "software for few". They pointed to tiny communities, internal tools, or one-off client workflows that can now get purpose-built software instead of being forced into generic SaaS. The sharpest skepticism was about confusing tool-building with actual progress. Some people saw these apps as sophisticated procrastination, especially in health, learning, and parenting. Others thought the article’s examples were over-optimized lifestyle management where a note, spreadsheet, script, or walk outside would do. But even that criticism clarified the dividing line. This is compelling when custom software removes friction or gives you control that packaged software with ads, subscriptions, or surveillance does not. It is much less compelling when the app becomes the hobby and displaces the thing it was supposed to support. A second practical theme was that today’s platforms are still hostile to this future. Commenters complained less about model quality than about distribution, hosting, and mobile operating systems. iOS in particular was treated as structurally opposed to personal software, because running your own app still means code-signing hoops, developer fees, or settling for a Progressive Web App. On the infrastructure side, many thought the article’s hosted stack was absurdly expensive for tiny personal tools. The consensus was simple. If the app is for one person, use the lightest thing that can possibly work, often SQLite on a cheap VPS, a Raspberry Pi, local-first storage, or even plain JSON files. In other words, LLMs may have made creation cheap, but the rest of the stack still needs to catch up.