Ethan Mollick

The AI landscape has fundamentally shifted from back-and-forth chatbot conversations to agentic systems that can independently complete multi-step tasks using tools, requiring users to understand three distinct components: Models, Apps, and Harnesses.

“Author of "A Guide to Which AI to Use in the Agentic Era" in One Useful Thing”

AI exhibits a 'jagged frontier' where it excels at some tasks while struggling with others in ways that don't align with human intuition about task difficulty, and this jaggedness may persist even as overall AI capabilities improve, creating complementary work opportunities with humans.

“Author of "The Shape of AI: Jaggedness, Bottlenecks, and Salients" in One Useful Thing”

A practical guide recommending users choose between free AI tools for casual use and paid $20/month advanced systems from Anthropic, Google, or OpenAI based on their specific needs and use patterns.

“Author of "An Opinionated Guide to Using AI" in One Useful Thing”

Recent advances in context window size (Gemini 1.5's 1M+ tokens) and response speed (Groq's hardware) fundamentally expand AI's practical capabilities for real-world applications like document analysis and rapid instruction following.

“Author of "Thinking through opponent moves"”

AI has crossed a threshold where it can perform real, economically relevant work comparable to human experts on individual tasks, though not yet ready to replace entire jobs due to jagged capabilities across complex workflows.

“Author of "Real AI Agents and Real Work"”

Mass Intelligence is emerging as powerful AI models become increasingly accessible to ordinary users through improved interfaces, model routing, and free access to advanced reasoning capabilities previously restricted to paying customers.

“Author of "Mass Intelligence" in One Useful Thing”

The Pentagon's demand for blanket AI approval from companies like Anthropic without transparency, combined with threats to label them supply chain risks, creates dangerous misalignments over military AI use policies including autonomous weapons and domestic surveillance.

“Quoted on government and nonprofit adoption of transformational AI”

“Author of Co-Existence: The Next Phase of AI; discussed which skills should not be automated and how people can maintain expertise in an AI-dominated future.”

“Expressed impressedness with DeepMind's Genie 4 model capabilities.”

“Wrote article on prompt engineering and effective AI usage, cited in documentation as reference for agent prompting techniques.”

“Ethan Mollick interpreted the PHP2Shell moment as attack and defense accelerating beyond human response speeds.”

“Mollick wrote about the Bitter Lesson applied to organizational management, discussing how agents can self-organize without human design.”

“Author and AI researcher who writes about agent coordination and management, discussed how agents self-organize without human management structures.”

“Had early access to Astra and noted it does complex meaningful work autonomously for days.”

“Posted on Bluesky warning that cybersecurity is going to become a mess soon due to AI agents using internet in unintended ways.”

“Wrote about the Hugging Face Incident demonstrating AI agents' ability to self-organize and collaborate, and the concept of the Twilight Factory.”

“Wharton professor who studies AI tools; cited for analysis comparing ChatGPT Work's approach to Claude's Cowork.”

“Author discussing how organizations struggle when employees become dramatically more productive with AI, breaking existing coordination systems.”

“Gave hands-on impression of Kimi K3: very good open-weights model but not Sol Max or Fable.”

“Questioned whether Trump administration's China AI model restrictions are true national security measures or industrial policy.”

“Author discussing how work is changing along the exponential curve of AI capability improvements, with AIs increasingly used as agents rather than chatbots.”

“Created a Stream Deck plugin for controlling Codex tasks.”

“Evaluated Thinking Machines' Inkling model and found it underperforms compared to frontier Chinese open-weights models on the Lem Test.”

“Byline author analyzing performance gap between open-weight and closed models using AA-Briefcase scoring data.”

“Wharton AI expert advises young job seekers to focus on tasks rather than skills, arguing that expertise in specific tasks is what allows people to stay ahead of AI.”

“Byline author and researcher discussing cognitive impacts of AI use in learning and work, including papers from research team at Wharton examining AI's effects on student learning.”

“Tested Claude Fable 5 and wrote detailed analysis of its capabilities across various tasks, finding it represents a major leap in AI capabilities.”

“Tested Claude 5 Fable extensively and documented his experience working with the Mythos-class model on complex projects.”

“Wharton professor who tested Claude Fable 5 and noted it can run autonomous workflows for over 9 hours on a single design document, scoring 91/100 on senior engineer benchmark.”

“Author of 'Co-Intelligence' book (New York Times bestseller, translated to 25+ languages) who is writing a new book 'Co-Existence' about working with AI that is sometimes better than humans; discusses his experience working with AI while writing and building his book website using Claude.”

“Posted a bundle of Mythos/Fable outputs including stable browser games and travel visualization tools, and wrote about the experience in One Useful Thing essay.”

“Author of One Useful Thing newsletter providing detailed hands-on review and analysis of GPT-5.5 capabilities across coding, research, and creative tasks.”

“Author analyzing how AI interfaces lag behind model capabilities and impact on user productivity and accessibility.”

“Authored article discussing when AI should and shouldn't be used, covering practical wisdom on AI capabilities and limitations.”

“Analyzed exponential AI capability improvements and the transition from co-intelligence to AI management era.”

A field experimental study of 758 consultants examines how AI affects knowledge worker productivity and work quality, revealing a jagged technological frontier where performance gains are uneven across task types.

“Co-author of BCG/Harvard study examining AI effects on consultant productivity with 758 participants.”

Companies are capturing only modest gains from AI adoption despite individual workers reporting significant productivity improvements, because organizational innovation muscles have atrophied and each company must discover its own AI integration approach.

“Developed a leadership framework for AI adoption in companies, identifying gaps between AI performance gains and organizational capture.”

AI may solve platform degradation and competitive races-to-the-bottom by raising average user intelligence, enabling people to see through manipulation and demand genuine value.

“Wrote about 'mass intelligence' concept where nearly a billion people now have access to near-genius-level AI models.”