The Jagged Frontier: OpenAI Commits to Research Acceleration
OpenAI’s Aaron Chatterji maps "The Jagged Frontier" of AI agents, showing how surging success rates accelerate research despite uneven task performance.

TL;DR
- OpenAI AI agents completed nearly half of two-to-four-hour research tasks without human intervention by October 2026, up from 23% in March.
- Success rates for longer, four-to-eight-hour tasks increased from 13% to 35% during the same period.
- Agent performance varies significantly by task type, with higher success in communication and analysis compared to build and coding tasks.
- Humans are still considered a core part of the AI research loop, particularly for idea generation and research taste.
- OpenAI is collaborating with Microsoft and NVIDIA, with NVIDIA intending to invest up to US$100bn in the company.
- Epoch AI's R&D taxonomy is used to classify agent tasks, and OpenAI uses an internal version of the Epoch Capabilities Index for comparison.