- The “Jagged Frontier”: AI’s capability across tasks is uneven, not uniformly good or bad. Mollick pictures it as an invisible fortress wall — everything inside is easy for the AI, everything outside is hard — where two tasks that look equally difficult to a person can sit on opposite sides of the wall (writing a sonnet vs. writing an exactly fifty-word poem) [2]. One team of consultants found the wall the hard way: given a task purpose-built to sit outside it, consultants who used AI scored worse (60–70% correct) than those who worked without it (84% correct) [1].
- The first of Mollick’s four working rules is “always invite AI to the table”: since the frontier’s shape isn’t known in advance, the only way to find it for your own job is repeated, disciplined experimentation rather than a one-time verdict [1].
- The second rule, “be the human in the loop,” keeps a person accountable for the final judgment call even where AI drafts or analyzes most of the material [3].
- Mollick distinguishes two working postures: “centaurs” keep a strategic, clean division of labor — deciding which whole tasks go to the AI and which stay with the person, the way Mollick lets AI produce the graphs while he picks the statistical approach [4]; “cyborgs” blend the two together mid-task, moving back and forth across the Jagged Frontier within a single piece of work, the way Mollick asks AI to rewrite a stuck paragraph ten different ways and then picks pieces from what comes back [4].
Co-Intelligence: Living and Working with AI
Mollick, a Wharton professor who researches how AI reshapes work, argues for an augment-don't-surrender posture — always invite AI to the table, but stay the human in the loop — that Management Craft draws on for using AI to build observing expertise without outsourcing the definition of greatness.
- Author
- Published
- 2024
- Purchase
- Amazon
- You get one impressive answer from a model and start trusting it broadly: without this book, one good result reads as general competence. Mollick’s Jagged Frontier says capability is uneven task-by-task, so the one good answer tells you almost nothing about the next task.
- You’re deciding how to use AI while building Knowing what great looks like: without this book, you either avoid AI entirely or hand it the whole judgment call. Mollick’s rules say invite it in for the drafting and pattern-finding, but keep yourself as the human in the loop who sets the local standard of greatness — the model can’t do that part for you.
- You’re deciding how deeply to integrate AI into your own work: without this book, “using AI” reads as one thing. Mollick’s centaur/cyborg distinction gives you two different postures to choose between deliberately, rather than drifting into whichever one happens by default.
1