Ethan Mollick is an Associate Professor at the Wharton School, University of Pennsylvania, the author of Co-Intelligence: Living and Working with AI (2024), and the writer of the One Useful Thing Substack, one of the most widely read publications on practical AI adoption. His work spans AI capability research and the questions of how organizations adopt AI in practice.
Concepts and frameworks
Mollick coined the "jagged frontier" concept with collaborators in a BCG study: AI performance is excellent at some tasks and poor at others within the same domain, which makes it difficult to predict where AI will succeed or fail. He has continued to develop the idea, extending it in 2026 from capability distribution (the technical claim) to organizational deployment (the practical claim).
He distinguishes the AI model, the application wrapper, and the tool-use system in a "Models, Apps, Harnesses" framework, arguing that harnesses now matter more than differences between models. He documents the shift from chatbot-era AI to agentic AI, framing the change as a move from chatbots that say things to agents that do things (Three Years from GPT-3 to Gemini 3, Source: oneusefulthing.org).
Adoption guidance and benchmarking
Mollick writes detailed guides to AI tools that track the changing landscape, including A Guide to Which AI to Use in the Agentic Era. On July 23, 2026 he published an updated "opinionated guide" to choosing AI systems, arguing that agentic systems capable of hours of autonomous work have changed what "using AI" means for most tasks (Source: oneusefulthing.org). On evaluation, he advocates that organizations "interview" AI models on their actual tasks rather than relying on standardized benchmarks (Giving Your AI a Job Interview, Source: oneusefulthing.org).
Positions on enterprise deployment
"Nobody knows anything" (NYPL talk, May 2026)
In late May 2026, addressing several hundred corporate leaders at the New York Public Library, Mollick endorsed William Goldman's "nobody knows anything" as the honest read on the AI-deployment moment:
"I spend my time talking to AI labs, famous people, I talk to CEOs all the time, and nobody knows anything. We're all making this up as we go along. So anyone who's like, 'We have the playbook' — they're lying to you."
He frames the position as consistent with the jagged frontier observation but extended from capability distribution to organizational deployment. He poses two diagnostic questions:
"How good and how fast? How long does this exponential curve continue and at what point does it ease off and how sharp will it be? That determines everything else."
This position is anchored as foundational via Lichtenberg — 'Nobody knows anything' and 'this time is different': the phrases that define — and haunt — the AI economy (Fortune, Lichtenberg, May 25 2026).
KPIs and the IT department (Economist, April 2026)
In an April 1, 2026 Economist "By Invitation" piece, Mollick advanced a structural-organizational explanation for the apparent productivity gap, noting that Bank of America's own bull-case math implies a ceiling of roughly 0.66% but realizes only about 0.1% economy-wide:
"KPIs are the biggest enemy at this point. They force you into very bad paths in the experimentation phase. The very nature of saying we need a 10% improvement constrains the kind of use cases that you see."
His argument is that the most significant AI applications cannot be produced through KPI-driven targets because they replace processes rather than improve them, and that the IT department's risk-reduction mandate is structurally hostile to experimentation that crosses process boundaries. Mollick characterizes this as a rational-conservatism reading of slow enterprise deployment, which he distinguishes from irrational-exuberance and technological-inadequacy framings (Source: economist.com).
The "AI consulting arms" observation
Mollick points to AI labs building their own deployment consulting arms as bear-side structural evidence:
"It's weird that the AI companies are all now building their own consulting arms to do AI deployment. If the models are so good that you think they're going to destroy all white-collar jobs, shouldn't they also be able to help you deploy systems?"
He pairs the framing with labor-disruption timelines: if AI can replace white-collar workers wholesale, AI itself should be able to deploy AI, so the persistence of human consulting layers at the labs (for example, Anthropic Solutions, and OpenAI partnerships with McKinsey and BCG) is, on his reading, evidence that the playbook does not yet exist even at the labs.
Relationships
- supports: AI Bubble Debate (KPI critique and consulting-arms observation are bear-side structural anchors)
- supports: Jagged Frontier (originator concept; extended in 2026 to organizational deployment)
- related: Lichtenberg — 'Nobody knows anything' and 'this time is different': the phrases that define — and haunt — the AI economy — primary anchor for the May 2026 NYPL public position
- related: So, About That AI Bubble — Rogé Karma (The Atlantic, May 2026) — Karma cites Mollick's "chatbots that say things → agents that do things" framing as part of the burden-of-proof inflection
- related: AI Labor Disruption — KPI brake reframes displacement timelines
Sources in Wiki
- Lichtenberg — 'Nobody knows anything' and 'this time is different': the phrases that define — and haunt — the AI economy (Fortune, Lichtenberg, 2026-05-25)
- A Guide to Which AI to Use in the Agentic Era (2026-02-17)
- Three Years from GPT-3 to Gemini 3 (Source: oneusefulthing.org) (2025-11-18)
- Giving Your AI a Job Interview (Source: oneusefulthing.org) (2025-11-11)
- The IT department is where AI goes to die, The Economist, By Invitation, 2026-04-01 (Source: economist.com)