November 23–27, 2026
Founder
Seldon AI
The emergence of System One and specialized decision models promises dramatically faster and cheaper AI inference. Yet early production experience reveals a more complicated reality: replacing expensive LLM calls with specialized models can introduce behavioral regressions, reduce agent reliability, and even increase end-to-end costs. This talk surveys the rapidly evolving decision-model landscape, examining real-world successes, failures, and the technical challenges of production adoption. We explore why effective optimization often requires refactoring entire AI processing pipelines rather than simply replacing individual models, and discuss the emerging opportunity for automated, continual optimization across specialized models, LLMs, and deterministic software. Attendees will gain practical insights into adopting decision models, evaluating tradeoffs, and designing more efficient and reliable AI systems.
See every session across the AI Agents Conference November 2026 line-up.
View Agenda