Ethan Mollick's Evolving Guide to AI Tools: From Chatbots to Agentic Systems
Ethan Mollick's guide to AI tools has significantly evolved over the past year, shifting focus from traditional chatbots to more advanced agentic systems. Initially, the landscape was dominated by conversational AIs like ChatGPT, Claude, and Gemini, with specific models such as Claude 4 Opus and Gemini 2.5 Pro being prominent. Deep Research was also noted as a valuable alternative mode. The current iteration of Mollick's guide highlights agentic systems, which are designed to perform extensive human-like work autonomously in a single operation. Google's Gemini has been removed from the latest list due to a lack of a clear offering in the 'Codex/ChatGPT Work/Cowork' category. Mollick explains how users can grant AI access to their computers through specific modes within the ChatGPT and Claude applications. ChatGPT offers 'Work' and 'Codex' modes, while Claude provides 'Cowork' and 'Code' modes. These desktop app modes are distinct from their mobile counterparts, offering enhanced capabilities due to their ability to access the user's computer. Notably, enabling 'Work' mode on the ChatGPT mobile app allows its Code Interpreter to access the internet, a feature not available in the standard 'Chat' mode.
AI tool recommendations are rapidly changing, reflecting the swift advancement from basic conversational agents to more sophisticated autonomous systems. The distinction between different AI modes, particularly those granting computer access, can be confusing due to inconsistent naming conventions across platforms. This highlights a user experience challenge in navigating increasingly complex AI functionalities. As AI agents become more capable of independent task execution, questions arise regarding data privacy, security, and the potential for unintended consequences when granting broad system access. The evolution suggests a future where AI acts less as a tool and more as a digital collaborator, necessitating clear user controls and transparent operational frameworks.
AI-generated to prompt reflection — not editorial opinion, not advice, not a statement of fact. How this works.