An opinionated guide to which AI to use to do stuff

2026-07-27 · Source: Simon Willison's Weblog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, quick

Summary

Ethan Mollick's "opinionated guide to which AI to use to do stuff" has evolved significantly, shifting focus from chat-centric models like ChatGPT, Claude, and Gemini to more powerful "agentic systems" capable of extensive human-equivalent work. Notably, Gemini has been removed from Mollick's current recommendations due to Google's lack of a proven entry in the "Codex/ChatGPT Work/Cowork" category, with Gemini Spark still unproven. The guide also clarifies how to grant AI access to computers, distinguishing between AI company-provided environments (e.g., ChatGPT Work, Claude Cowork) and direct access via desktop applications (ChatGPT's Work/Codex, Claude's Cowork/Code), which offer enhanced capabilities. A key distinction is that ChatGPT mobile's "Work" mode enables its Code Interpreter to access the internet, unlike its desktop counterpart.

Key takeaway

For AI Engineers evaluating current large language model capabilities, understanding the shift towards agentic systems is crucial for task automation. You should prioritize exploring platforms like ChatGPT Work or Claude Cowork for complex, multi-hour tasks, and consider desktop applications for direct AI computer access to leverage their full feature sets. Be aware of the unintuitive naming conventions and distinct capabilities, such as ChatGPT mobile's internet-enabled Code Interpreter in "Work" mode, to optimize your workflow.

Key insights

AI usage is shifting from simple chat to agentic systems that perform complex, multi-hour tasks.

Principles

• Agentic AI systems automate extensive human work.
• AI access modes vary significantly by platform.
• Naming conventions for AI tools are often confusing.

In practice

• Explore ChatGPT Work or Claude Cowork for agentic tasks.
• Use desktop apps for enhanced AI computer access.
• Utilize ChatGPT mobile's Work mode for internet-enabled Code Interpreter.

Topics

• Agentic AI
• Large Language Models
• ChatGPT Work
• Claude Cowork
• AI Access Modes
• Code Interpreter

Best for: AI Engineer, Software Engineer, Consultant

Related on AIssential

• ChatGPT is now a partner for your most ambitious work — ChatGPT Work is an agent powered by GPT-5.6 that automates multi-step tasks across apps, files, and web, transforming goals into finished work.
• I use Anthropic's Claude AI tools for very different jobs: How to pick between models, Code, and Cowork — Anthropic's AI tools distinguish between chatbots for conversation and agents for complex, multi-step task execution.
• From agents to world models: What San Francisco revealed about AI’s next phase — AI is transitioning from linguistic models to agentic systems that act, use tools, and comprehend physical and virtual environments.
• 😺 LIVE now: Agents for total beginners — AI agents and automation simplify repetitive tasks, freeing human judgment for complex work.
• Empowering Users to Get More Done With Agent Mode — Agent Mode automates complex, multi-step tasks using integrated tools, shifting AI interaction from chat to agentic workflows.
• Claude Cowork expands to mobile and web — AI agents are expanding beyond coding into general knowledge work, becoming cross-device administrative assistants.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Simon Willison's Weblog.