AI agents gain software structure for improved reliability
Agents as Software: A Programming Languages Agenda for Agent Reliability
Programming LanguagesArtificial Intelligence
Summary
AI agents today act like complex software, using tools and making decisions that affect the real world. The authors point out that these agents are tricky to test and fix because their 'programs' are spread across many parts like prompts and memories. They propose treating agents like software programs that can be checked, monitored, and improved systematically. This approach doesn't make AI agents fully predictable but helps control and repair their behavior more effectively.
What this means in practice
- •For software developers: Design AI agents with better tools to test and monitor their behavior systematically before and during deployment.
- •For ai operations teams: Implement monitoring and repair workflows to improve the reliability of deployed AI agents based on observed failures.
A position paper. It proposes an approach and reports no results.
Authors
Shraddha Barke, Adithya Murali
Abstract
AI agents increasingly resemble software systems: they call tools, remember facts, follow policies, delegate work, and take actions with real consequences. % Yet the ``program'' of an agent is scattered across prompts, tools, memories, workflows, and execution traces, making its behavior difficult to inspect through ordinary testing and debugging alone. % This essay argues that a programming-systems perspective offers a natural lens for making agents reliable. % We recast agents as programmable artifacts whose behavior can be specified over traces and state, checked before deployment, monitored during execution, and improved from observed failures. % The goal is not to make probabilistic agents behave like deterministic programs, but to give them enough structure that their behavior can be reasoned about, controlled, and repaired.