What Do Production AI Agents Actually Look Like? An Empirical Study скачать в хорошем качестве

What Do Production AI Agents Actually Look Like? An Empirical Study

A deep dive into a comprehensive empirical study of AI agents deployed in production environments. This paper surveyed ~300 practitioners and conducted in-depth interviews to understand what real-world AI agent deployments look like. Key findings: Most deployed agents are workflows rather than fully autonomous systems Organizations build custom agent loops rather than using frameworks like LangChain Reliability remains the #1 challenge Heavy reliance on human expert evaluation over formal benchmarks Paper: https://arxiv.org/abs/2512.04123v1 0:00 Introduction 0:48 Why build agents? 1:44 Agent properties and deployment 3:38 Agent architectures in production 5:18 Multi-agent systems 6:21 Evaluation challenges 10:14 Reliability as the key challenge 11:38 Conclusion    • Coding Agents Are Outliers      • Claude Skills will eat (most) agents