An opinionated guide to which AI to use to do stuff

Artificial intelligence has shifted from basic chatbots to agentic systems that use virtual or local computers to plan, execute, and complete hours of human work in minutes. For routine, low-stakes questions, free default chatbots are adequate, but complex or high-stakes matters require advanced models like Claude’s Opus and Fable or ChatGPT’s GPT-5.6 Sol set to high thinking levels.

For real tasks, users generally choose between ChatGPT and Claude starting at twenty dollars a month. These systems offer two main setups: using a virtual computer provided by the AI company, or giving the AI access to the user's own computer through dedicated apps. Cloud-based modes handle emails, research, and scheduling, while local computer access allows AIs to handle complex multi-file projects, check book references, or directly operate software like Blender. Users should maintain strict approval settings for sending emails or spending money to guard against errors and prompt injection security risks.

Other platforms offer specific strengths. Microsoft Copilot lags in agent abilities, and Google lacks a leading frontier agent system, but Google excels in research via Notebook and handles direct video editing through Gemini Omni, while ChatGPT features advanced native voice modes and image generation. Ultimately, working with these modern agents resembles management rather than chatting, requiring users to delegate tasks, review outputs, and provide corrections much like managing a human team.