The article identifies three main types of AI products that are currently effective: chatbots, completions, and agents. Chatbots, while popular, are often limited to generic interactions because users can find better options like ChatGPT. Many new products simply repackage chatbots, adding little value. Explicit roleplay chatbots tap into niche markets like adult content, but they face competition from major AI labs that are likely to expand their offerings in that area. Chatbots with tools, labeled as AI assistants, struggle because users can easily manipulate them, making it hard to integrate real support functions. Completions-based products, exemplified by GitHub Copilot, suggest code snippets as users type, enhancing existing workflows without requiring conversation. These have proven successful in coding but haven’t gained similar traction in other areas like professional writing. The article questions why more industries haven’t adopted this model, suggesting that familiarity with autocomplete in coding may play a significant role. The third category, agents, emerged with advancements in AI technology, allowing users to input a request and let the model carry out the task autonomously. This contrasts with chatbots that require ongoing interaction. Coding agents are particularly effective due to the ease of verifying code changes. The potential for agents extends beyond coding to tasks like research, where models can quickly sift through data. AI-generated feeds are on the horizon, with companies experimenting with personalized content across platforms, though none have gained significant traction yet.