Audiobook Summary and Review by StoryShots
Same tool, same session, brilliant on one task and confidently wrong on the next.
An AI that hallucinates fake citations with total confidence is the same AI that, when used by consultants in a controlled study, helped them dramatically outperform peers on creative tasks.
That contradiction is the starting point of Co-Intelligence: Living and Working with AI, by Wharton professor Ethan Mollick.
Generative AI, he shows, is not a better calculator.
It is something we have never built before.
Most people open ChatGPT expecting it to behave like Excel.
Type the right formula, get the right answer, every time.
That expectation breaks immediately.
Generative AI predicts the next word based on probability, not stored fact.
Ask it the same question twice and you can get two different answers, one brilliant, one fabricated, delivered with identical confidence.
Every time you treat an AI output as a fact instead of a first draft, you are gambling with a system that was never built to know the difference.
Understanding this changes how you interact with these tools daily, and it raises the question of where their abilities actually run out.
Ask an AI to write a sonnet and it delivers something genuinely moving.
Ask it to count the words in that sonnet and it gets the number wrong.
This is the jagged frontier, an invisible, shifting boundary where a machine is superhuman at one task and incompetent at the task right next to it.
There is no manual for where this line sits in your job.
The only way to find the edge is to push against it yourself, with your own work.
The frontier does not care about your job title.
It only cares about the task in front of you.
That instability is exactly what the next idea starts to resolve.
In a Boston Consulting Group study, consultants using GPT-4 did not just improve.
Many performed at a level that beat the vast majority of their peers on tasks squarely inside the frontier.
Outside it, the same consultants using the same tool did worse than those who ignored it completely.
The resolution is not trust AI more or trust it less.
Expertise and AI multiply each other.
People who already knew their field spotted a wrong answer instantly and steered around it.
People without that expertise could not tell brilliance from fabrication, and the AI happily handed them both.
Domain knowledge just became the most valuable insurance policy against a machine that lies with total confidence.
That leaves a bigger question this trailer cannot fully settle: what happens to organizations built entirely around human hierarchies once the smartest voice in the room might be software.
If this changed how you think about using AI at work, someone on your team would probably get real value from hearing it too.
This summary of Co-Intelligence threads together AI's unpredictable, human-like behavior, the shifting jagged frontier that defines where it helps and where it fails, and the discovery that expertise, not tech skill, determines who benefits from it.
The full book goes further, into four rules for working with AI daily, the difference between centaur and cyborg collaboration styles, and four competing scenarios for where this technology takes work and society next.
It also unpacks the alignment problem, why AI might not share human ethics at all, using real examples from Ethan Mollick's own classroom and consulting research.
For the full summary of Co-Intelligence, including the infographic and animated video breakdown, head to the StoryShots app.