# Ethan Mollick’s guide to choosing AI tools for work [Canonical reading](https://escapemeridian.com/readings/ethan-mollick-s-guide-to-choosing-ai-tools-for-work-73cfeaa7) Ethan Mollick published this guide on July 23, 2026, arguing that AI use is shifting from chatbot conversations toward agentic systems that can use tools and a computer to carry out tasks. He recommends Claude or ChatGPT for many people seeking AI help with substantial work, while distinguishing company-provided virtual computers from access to a user’s own computer. His examples include agents researching email, preparing teaching materials, reviewing a book manuscript, and operating software; these are his reported experiences, not evidence that every user will get similar results. Mollick warns that permissions matter: agents may send messages or change files, and prompt injection—text designed to manipulate an AI—remains a risk. He advises limiting access and requiring approval before actions that send, spend, or delete. His recommendations also distinguish tools for particular needs, including Google’s tools for research and video, and tools with image or voice features. How often do these agents make consequential mistakes when given access to email or files, and how much do approval settings reduce that risk? Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims. ## Original source One Useful Thing · Article Original source date: Jul 23, 2026 (2026-07-23T18:05:24.000Z). Date basis: Date supplied by the publisher’s feed. Added to Escape Meridian: 2026-10-29T17:48:04.225Z. First observed: 2026-09-23T05:31:19.956Z. Retrieved: 2026-10-29T12:45:43.423Z. - [An opinionated guide to which AI to use to do stuff](https://www.oneusefulthing.org/p/an-opinionated-guide-to-which-ai-b22) ## Limits Text as retrieved on October 29, 2026; originally published July 23, 2026. Mollick’s examples and assessments reflect his own use and judgment; they do not establish how these tools perform for other users or tasks. The recommendations describe products and capabilities as Mollick presented them in the article; the text does not provide comparative testing or independent evidence of reliability. ## Open question How often do these agents make consequential mistakes when given access to email or files, and how much do approval settings reduce that risk? ## Source-stated Mollick says AI use is expanding from chatbot conversations to agentic systems that can use tools to plan and act. > Now, it means using an agentic system, where the AI is capable of doing the equivalent of many hours of real human work in one go by combining the brains of an AI model with a set of tools that let it plan and act for you. - [Supporting source](https://www.oneusefulthing.org/p/an-opinionated-guide-to-which-ai-b22) ## Source-stated Mollick recommends Claude or ChatGPT as the main choices for many people seeking AI help with substantial work. > There are only two choices for most people who want to get the most out of AI right now: ChatGPT or Claude - [Supporting source](https://www.oneusefulthing.org/p/an-opinionated-guide-to-which-ai-b22) ## Source-stated Mollick describes an agent sending an email after he had previously granted it permission, and advises keeping approval settings on for consequential actions. > I had previously given ChatGPT permission to send email on my behalf, and Claude was told to ask me first. - [Supporting source](https://www.oneusefulthing.org/p/an-opinionated-guide-to-which-ai-b22) ## Source-stated Mollick says prompt injection remains an unresolved risk for agents that read email and browse the web. > The AI labs are working on this problem, and models have gotten more resistant, but it is not solved. - [Supporting source](https://www.oneusefulthing.org/p/an-opinionated-guide-to-which-ai-b22) ## Source-stated Mollick reports that an AI reviewed his book manuscript and that he found its notes accurate, while cautioning that human judgment remained useful. > One sign of how far AIs have come is that every one of the AI's notes was accurate and there were no hallucinated page numbers, no invented text, no errors I could spot at all. - [Supporting source](https://www.oneusefulthing.org/p/an-opinionated-guide-to-which-ai-b22) ## Source and contributor context One Useful Thing · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media retrieval status: unchecked. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Agentic AI AI systems that plan and carry out a series of steps toward a goal, often by using other software, rather than giving one answer. - [Definition reference](https://developers.google.com/machine-learning/glossary#agentic) ## Prompt injection Hidden instructions placed in a document, email or web page that try to take control of an AI system reading it. - [Definition reference](https://en.wikipedia.org/wiki/Prompt_injection) ## Ethan Mollick Ethan Mollick is a Wharton researcher and author exploring how AI changes work, learning and decision-making. - [Explore this context](/entities/ethan-mollick) - [Reference](https://mgmt.wharton.upenn.edu/profile/emollick/) - [Reference](https://www.oneusefulthing.org/) ## One Useful Thing One Useful Thing is Ethan Mollick’s publication about AI in work, education and everyday practice. - [Explore this context](/entities/one-useful-thing) - [Reference](https://www.oneusefulthing.org/) ## Claude Claude is Anthropic’s family of AI models and assistants for tasks such as writing, analysis and coding. - [Explore this context](/entities/claude) - [Reference](https://claude.com/product/overview) - [Reference](https://www.anthropic.com/news) - [Reference](https://www.anthropic.com/research/team/economics) ## Editorial provenance Version 1. Publication ID: ced2d962-ac42-4be5-3397-3792228368bf. Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5. Source check: passed by anthropic/claude-opus-5.5 on 2026-10-29T17:46:50.061Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: bridge\_admitted\_by\_model, second\_opinion\_categorical. Observation: a3e691bf359dfd03cbd001c2043051b5bf04fe42a248393188966804f574823f. Source text hash: 17a5038d542a16160122db58136fb4c9066252af321064b300bfe85533d98b1d. ## Revision and source-check history ### Version 1 — 2026-10-29T17:48:04.225Z Ethan Mollick published this guide on July 23, 2026, arguing that AI use is shifting from chatbot conversations toward agentic systems that can use tools and a computer to carry out tasks. He recommends Claude or ChatGPT for many people seeking AI help with substantial work, while distinguishing company-provided virtual computers from access to a user’s own computer. His examples include agents researching email, preparing teaching materials, reviewing a book manuscript, and operating software; these are his reported experiences, not evidence that every user will get similar results. Mollick warns that permissions matter: agents may send messages or change files, and prompt injection—text designed to manipulate an AI—remains a risk. He advises limiting access and requiring approval before actions that send, spend, or delete. His recommendations also distinguish tools for particular needs, including Google’s tools for research and video, and tools with image or voice features. How often do these agents make consequential mistakes when given access to email or files, and how much do approval settings reduce that risk? ## Original source One Useful Thing · Article Original source date: Jul 23, 2026 (2026-07-23T18:05:24.000Z). Date basis: Date supplied by the publisher’s feed. Added to Escape Meridian: 2026-10-29T17:48:04.225Z. First observed: 2026-09-23T05:31:19.956Z. Retrieved: 2026-10-29T12:45:43.423Z. - [An opinionated guide to which AI to use to do stuff](https://www.oneusefulthing.org/p/an-opinionated-guide-to-which-ai-b22) ## Limits Text as retrieved on October 29, 2026; originally published July 23, 2026. Mollick’s examples and assessments reflect his own use and judgment; they do not establish how these tools perform for other users or tasks. The recommendations describe products and capabilities as Mollick presented them in the article; the text does not provide comparative testing or independent evidence of reliability. ## Open question How often do these agents make consequential mistakes when given access to email or files, and how much do approval settings reduce that risk? ## Source-stated Mollick says AI use is expanding from chatbot conversations to agentic systems that can use tools to plan and act. > Now, it means using an agentic system, where the AI is capable of doing the equivalent of many hours of real human work in one go by combining the brains of an AI model with a set of tools that let it plan and act for you. - [Supporting source](https://www.oneusefulthing.org/p/an-opinionated-guide-to-which-ai-b22) ## Source-stated Mollick recommends Claude or ChatGPT as the main choices for many people seeking AI help with substantial work. > There are only two choices for most people who want to get the most out of AI right now: ChatGPT or Claude - [Supporting source](https://www.oneusefulthing.org/p/an-opinionated-guide-to-which-ai-b22) ## Source-stated Mollick describes an agent sending an email after he had previously granted it permission, and advises keeping approval settings on for consequential actions. > I had previously given ChatGPT permission to send email on my behalf, and Claude was told to ask me first. - [Supporting source](https://www.oneusefulthing.org/p/an-opinionated-guide-to-which-ai-b22) ## Source-stated Mollick says prompt injection remains an unresolved risk for agents that read email and browse the web. > The AI labs are working on this problem, and models have gotten more resistant, but it is not solved. - [Supporting source](https://www.oneusefulthing.org/p/an-opinionated-guide-to-which-ai-b22) ## Source-stated Mollick reports that an AI reviewed his book manuscript and that he found its notes accurate, while cautioning that human judgment remained useful. > One sign of how far AIs have come is that every one of the AI's notes was accurate and there were no hallucinated page numbers, no invented text, no errors I could spot at all. - [Supporting source](https://www.oneusefulthing.org/p/an-opinionated-guide-to-which-ai-b22) ## Source and contributor context One Useful Thing · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media retrieval status: unchecked. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Agentic AI AI systems that plan and carry out a series of steps toward a goal, often by using other software, rather than giving one answer. - [Definition reference](https://developers.google.com/machine-learning/glossary#agentic) ## Prompt injection Hidden instructions placed in a document, email or web page that try to take control of an AI system reading it. - [Definition reference](https://en.wikipedia.org/wiki/Prompt_injection) ## Ethan Mollick Ethan Mollick is a Wharton researcher and author exploring how AI changes work, learning and decision-making. - [Explore this context](/entities/ethan-mollick) - [Reference](https://mgmt.wharton.upenn.edu/profile/emollick/) - [Reference](https://www.oneusefulthing.org/) ## One Useful Thing One Useful Thing is Ethan Mollick’s publication about AI in work, education and everyday practice. - [Explore this context](/entities/one-useful-thing) - [Reference](https://www.oneusefulthing.org/) ## Claude Claude is Anthropic’s family of AI models and assistants for tasks such as writing, analysis and coding. - [Explore this context](/entities/claude) - [Reference](https://claude.com/product/overview) - [Reference](https://www.anthropic.com/news) - [Reference](https://www.anthropic.com/research/team/economics) ## Editorial provenance Version 1. Publication ID: ced2d962-ac42-4be5-3397-3792228368bf. Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5. Source check: passed by anthropic/claude-opus-5.5 on 2026-10-29T17:46:50.061Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: bridge\_admitted\_by\_model, second\_opinion\_categorical. Observation: a3e691bf359dfd03cbd001c2043051b5bf04fe42a248393188966804f574823f. Source text hash: 17a5038d542a16160122db58136fb4c9066252af321064b300bfe85533d98b1d. ### passed — 2026-10-29T17:46:50.061Z — anthropic/claude-opus-5.5 Publication: ced2d962-ac42-4be5-3397-3792228368bf attribution — pass: The guide is correctly attributed to Ethan Mollick, the byline author on One Useful Thing, and its recommendations are presented as his views and experiences. > An opinionated guide to which AI to use to do stuff The Summer 2026 Edition Ethan Mollick summary — pass: The summary accurately captures the shift to agentic systems, the recommendation of Claude or ChatGPT, the two ways of giving an agent a computer, and the warnings about permissions and prompt injection. It explains key terms in plain language. The question follows from the source's own caveats about permissions and unsolved prompt injection. > Basically, an agentic system gives an AI a computer to use. date — pass: July 23, 2026 matches the source dateline and the UTC day of originalPublished. modifiedAt equals the publication instant, so no later edit is indicated. This is a publisher page, so no archive disclosure is needed. The reading presents nothing from after publication and makes no hindsight claims, including about the forthcoming book. > Ethan Mollick Jul 23, 2026 support — pass: Each claim is source-stated and backed by an exact excerpt: agentic systems, two main choices, the email-permission episode, prompt injection being unsolved, and accurate manuscript notes alongside his use of human judgment. The reading hedges his examples as personal experience and does not overstate their reliability. > This is another reason to limit what your agent can touch, and to keep approval settings on for anything that sends, spends, or deletes. media — pass: No media was selected. No verified media candidates were supplied, so leaving media empty is appropriate. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/73cfeaa7efb67e03d4f40498fbdaaa681cf4a626f8862e31a96f2bddcb48eff6/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/ethan-mollick-s-guide-to-choosing-ai-tools-for-work-73cfeaa7.json)