Cursor Shares Research on Long-Running Agents: GPT-5.2 'So Much Better'
Cursor's research lead shared early findings on how agents behave in extended sessions, with GPT-5.2 showing marked improvement — a signal that matters for anyone building autonomous coding workflows.
@leerob from Cursor shared what he called "some cool things from our research on long-running agents," with a specific callout that GPT-5.2 is "so much better" for these use cases. While the post was light on specifics, the signal is significant: Cursor is one of the highest-volume consumers of frontier language models for code generation, and their internal benchmarks on agent persistence and coherence carry weight.
Unlock the full briefing
Get every story in today's briefing, the full archive, and the daily AI intelligence brief.
All stories today
Full archive
Daily brief
Cancel anytime. Payments powered by Stripe.