When Better Turns Do Not Make Better Agents: Diagnosing the Gap Between Next-Turn Metrics and Workflow Success

What changed

Supervised fine-tuning improves next-turn metrics in agent models, but these gains do not transfer to holistic autonomous workflow execution success.

Why you should care

Next-turn metrics are insufficient proxies for end-to-end task completion in AI agents.

Your move

Watch. Monitor shifts in benchmark standards toward holistic workflow evaluations.

Evidence

Preliminaryarxiv.org ↗

What to watch next

Adoption of standardized holistic workflow benchmarks across agent model evaluations.

Event

research

Event date

Sep 21, 2026

Relevant to

General AI readers