
Why AI writing reads as AI

Why AI is going vertical (again) | Dianne Penn (Anthropic)
Topic 12 of 18
Published Jul 26, 2026 · Clip 1:07:06–1:11:41 (4:35)
The pair discusses why AI writing still reads as AI. Diane says it depends on the goal: there are very active efforts on her team and on the research side to make Claude write better, and it should be clear where an idea is being led — for something like a monthly business review she'd happily have it written end-to-end by Claude. What matters more than who writes is verifiability: who verifies and who signs off. On why writing lags, she points to underinvestment in training for it and the jagged frontier — leaps in agentic behavior made writing and tone the rough edges — so the team is now focusing and prioritizing training for writing. She and the host close with the airplane Wi-Fi joke about human nature: once a capability jump lands, we immediately expect the next improvement, with proactivity likely next after tone and character.