An opinionated guide to which AI to use to do stuff
One Useful Thing · Ethan Mollick · 2026-07-23
Ethan Mollick recommends ChatGPT and Claude as the only viable agentic AI platforms for real work at $20/month, finding Google Gemini and Microsoft Copilot significantly behind, while warning users about prompt injection and permission risks.
Extraction
Topics: agentic-aillm-comparisonai-productivitychatgptclaude
Claims
- Agentic AI systems that give models access to a computer have created a qualitative shift, enabling AI to autonomously perform hours of human work in a single session.
- ChatGPT and Claude are the only viable general-purpose agentic AI platforms at $20/month; Google Gemini and Microsoft Copilot lag significantly in agentic capability.
- Prompt injection—where malicious text in emails or web pages hijacks an AI agent—remains an unresolved security risk that users should mitigate by limiting agent permissions.
- The skill required to use AI effectively has shifted from prompt engineering expertise to people-management skills such as delegation, feedback, and correction.
- For high-stakes decisions, users should access frontier thinking models (Claude Opus or Fable, GPT-5.6 Sol) rather than default free-tier models, which have meaningfully higher error rates.
Key quotes
Basically, an agentic system gives an AI a computer to use.
Until you trust the system (and understand its mistakes), leave everything to ask for approval first, which is the default. This also protects against a second risk, called prompt injection.
You don't need to be good at prompting, but rather at asking for what you want and correcting the AI when it doesn't get your intentions.