Ethan Mollick's *Co-Intelligence* is the book practitioners keep pointing to two years on — and the reason is one reframe and two ideas that actually hold up. This episode breaks down why treating AI as a collaborator rather than a search box changes the output, what the jagged frontier means for knowing when to trust it, and what the evidence underneath the framework really says.

AI-generated (NotebookLM) audio overview. Source: HexLocal in-house research — Research - Co-Intelligence (Ethan Mollick), AI Foundations Deep Dive (Dr. Priya Nair). Primary sources include Mollick's Co-Intelligence (Portfolio/Penguin, 2024) and the Harvard Business School / BCG field experiment published in Organization Science (2025).

- The co-intelligence reframe: treating AI as a collaborator rather than a search box or vending machine changes what you get out of it — and the evidence backs that up

- The jagged frontier: AI is superhuman on some tasks and surprisingly bad on adjacent ones, with no boundary that maps to human intuitions about difficulty

- The 758-consultant experiment: on the right tasks, AI lifted quality and speed; on a deliberately wrong task, it made consultants 19 points more likely to be wrong

- Mollick's four rules: always invite AI to the table, be the human in the loop, treat it like a person but tell it what kind, and assume this is the worst AI you'll ever use

- Where the framework holds and where it doesn't: strongest for operators figuring out first moves; thinner on power, governance, and what happens at scale

- The honest context: published 2024, its value is durability — it's the established practitioner standard, not a fresh arrival