Only three kinds of AI products actually work

Analysis of successful AI product categories reveals three working models: chatbots (like ChatGPT), code completion tools (like GitHub Copilot), and coding agents (like Claude Code). Chatbots face competition from general-purpose models, while completion-based products succeed by integrating seamlessly into existing workflows. Agents work because they handle complex multi-step tasks autonomously. Two emerging categories show promise: AI-generated feeds and AI-enhanced video games, though neither has achieved mainstream success yet. Most AI products fail because they're just chatbots competing with superior general-purpose alternatives, or they use chat as an interface when traditional UI would be more efficient.