Ethan Mollick’s guide to which AI tools to use has noticeably changed over the past year. Where it once emphasized chat-first models — including ChatGPT, Claude and Gemini, with specific mentions of models like o3, Claude 4 Opus and Gemini 2.5 Pro and Deep Research as a useful alternative — it now focuses much more on agentic systems.
What Mollick means by agentic systems
Mollick uses the term “agentic” for systems that can perform the equivalent of many hours of human work in one go. These are not merely conversational language models but systems that can act, integrate with tools and execute multi-step workflows autonomously.
Which models were in focus previously?
A year ago, Mollick’s list centered on ChatGPT, Claude and Gemini, and he named o3, Claude 4 Opus and Gemini 2.5 Pro as notable models. He also noted Deep Research as a useful alternative mode.
What’s different now?
The current guide emphasizes systems that can act as agents — notably by giving the AI access to a computer so it can perform tasks directly. Mollick argues that the most powerful way to use AI is to grant it access to your computer.
Product modes and name confusion
Mollick explains how to give ChatGPT or Claude a computer to use and highlights the confusing naming across providers. For example:
- In ChatGPT, the relevant mode is called ChatGPT Work (and there are agent modes named Work and Codex in desktop apps).
- In Claude, analogous modes are called Cowork and Code.
He notes that these names do not map cleanly onto each other and that the Work and Cowork labels are not straightforward equivalents.
Mobile vs. desktop: the ChatGPT Work example
Mollick points out a striking and unintuitive difference between ChatGPT Work on mobile and ChatGPT Work inside the desktop app (where the desktop Work functions more like a user interface on top of Codex). Specifically, switching ChatGPT mobile from "Chat" to "Work" can yield a version where the Code Interpreter container is no longer prevented from accessing the internet. This difference in behavior between platforms is something Mollick finds especially confusing.
Gemini’s removal from the list
Mollick has dropped Gemini from his recommendations because, in his view, Google does not yet have a clear offering in the Codex/ChatGPT Work/Cowork category — there is no obvious, comparable product to fill that role.
Why this matters in practice
The shift in Mollick’s guide highlights that choosing AI tools today requires looking beyond chat capabilities. For productivity and automating complex workflows, it matters whether the AI can access files, run commands and integrate with the user’s computer. That capability is what Mollick identifies as the most powerful way to apply these systems.
Summary
Over the past year, Ethan Mollick’s guidance has moved from recommending chat-centric models toward advocating agentic systems that can execute extended work autonomously. He emphasizes the practical importance of giving AI systems computer access, and cautions that product names and platform behaviors vary in ways that can be confusing. Mollick also notes that Google’s Gemini does not currently offer a clear competitor in this specific agentic category.
