Writing on his newsletter One Useful Thing on October 20, 2026, Wharton professor Ethan Mollick announced the release of his second book on AI, 'Co-Existence,' 30 months after his first book, 'Co-Intelligence,' came out in early 2024. Mollick frames the gap between the two books as a way to measure AI progress, and illustrates that progress using the process by which he produced the article's charts. He writes that in the earlier 'era of chatbots,' he would have had to work 'back-and-forth with an AI in a chat window, guiding it through each new step.' For the new charts, he says he instead delegated the work to agentic AI, using 'Opus 5.5 in Claude Projects,' which he describes as letting a user talk to a single 'orchestrator' agent that spins off additional agents as needed. According to Mollick, he gave the AI 'a couple sentences of suggestions' and it handled research, graphing, layout, images and citations in minutes, work he says would have taken him hours. He adds that he gave 'handwavy reactions' such as 'make it better,' and that the AI caught a math error, noting he had originally planned to title the post '28 months on the exponential.' The account is a single author's first-person description, not an independent test.
- Mollick's second AI book, 'Co-Existence,' released October 20, 2026, 30 months after 'Co-Intelligence' (early 2024)
- He used 'Opus 5.5 in Claude Projects' with an 'orchestrator' agent that spins off sub-agents to produce the article's charts
- He reports the AI did research, graphing, layout, images and citations in minutes, and caught a math error in his draft
What it means for you
A respected observer is describing a practical shift: from AI you prompt step-by-step in a chat window, to AI you hand a task to and let it run. That's a genuine change in how some knowledge work — research, formatting, first drafts — can get done. But this is one person's anecdote about his own book promotion, not a controlled result, so treat it as a signpost, not proof. The real lesson for operations is the one Mollick stumbled into: the AI caught his math error, but he still needed to know the math was wrong to trust the fix.
Try this
Pick one repetitive 'gather-and-format' task your team does by hand this week — a report, a chart pack, a summary — and map the exact steps a person takes today. You'll learn whether the bottleneck is the tool or the fuzzy process underneath before you automate anything.
Who should care
Owners and managers whose teams spend real hours on research, reporting, and document assembly, and anyone weighing whether 'agentic' AI is meaningfully different from the chatbots they already tried.
Skip this if
You don't do much document or analysis production in-house, or you've already got a clear handle on where AI fits in your workflow — this is a trend-spotting essay, not a decision point.
Sources: One Useful Thing — read the original