AI is driving a boom in modestly successful software projects

AI coding tools are generating a measurable boom in small, moderately successful software projects rather than blockbuster apps. The author examined twenty different software marketplaces and plugin registries—Steam, Obsidian, Firefox extensions, Nexus mods, Blender extensions, and others—comparing successful projects before and after mid-2025 when agentic coding went mainstream. The pattern is consistent: roughly 2x growth in modestly successful projects (those with hundreds to thousands of users) but no growth in wildly successful software. For example, NVDA screen reader addons doubled in projects exceeding 1,000 downloads, HACS home automation plugins doubled above 100 installs, and Steam games with over 100 reviews jumped 60%. Meanwhile, massively successful products like macOS or Gmail haven't spawned new competitors. This suggests AI isn't failing—it's just solving problems at a different scale than people expected. The author deliberately focused on addons and plugins because they reveal actual usage through installation counts, sidestepping the noise of AI-generated repos that sit unused. More importantly, examining game mods and browser extensions shows whether AI helps solve ordinary problems people actually want solved, not just whether it can generate code. Game modders using AI to add multiplayer modes or new features to existing games aren't trying to revolutionize anything—they just want playable features, and AI appears to deliver that effectively. The author acknowledges that proving AI involvement is nearly impossible since hobbyist communities actively discourage people from disclosing AI usage due to stigma, but the 2x growth in successful projects against a 10x explosion in total projects strongly suggests AI is driving the difference. Big software companies aren't seeing explosive growth in new features because they already have the features they want and are technically constrained by their existing codebases. New features in mature products tend to be invisible—enterprise compatibility tweaks or internal reliability improvements that most users never notice. The author speculates that small, wildly successful projects might simply be rare regardless of AI, or they might take years to develop from thousands-of-users tools into hundred-thousands-of-users tools. The real takeaway: AI isn't producing fewer or worse software products. It's producing more software that serves niche audiences—thousands of small, useful projects instead of the next Gmail.