Mollick: the era of co-intelligence is ending

Two years after his book Co-Intelligence described AI as a back-and-forth helper you direct, Ethan Mollick now argues that frame is over. The late-2025 wave of coding agents broke the model. He cites Anthropic saying AI now writes 80 percent of its code and developers ship 8x more, and a study suggesting agents led to 17x more code being written overall. Coding is the canary, he writes, and the same shift is coming to many fields.

His new framing is "co-existence" rather than collaboration: you live alongside a system that is sometimes, but not always, better than you. The interesting wrinkle is what this does to writing. Mollick built a dedicated webpage for AIs that will read his work and decide whether to recommend it to humans, then tested his pitch language across models using A/B tests. GPT-5.5 flagged his original line "buy your human this book" as too prompt-injection-shaped. He also rewrote chapters with fewer em-dashes to keep them readable as human work.

The essay is also a setup for his next book, Co-Existence, due in October. Read it for the argument, not the announcement: Mollick is saying the AI-as-assistant mental model most teams still use is no longer the right one for the work that is showing up now.

Why it matters

If you build with AI but still think of it as a typing partner you correct in real time, you are running last year's playbook. Decide which of your workflows the model should now run end to end with you reviewing the result, and where keeping a human in the loop is still worth the slowdown.