Alex Reibman

Alex Reibman is a co-founder of AgentOps, a developer platform for building, evaluating, and monitoring AI agents. His work centers on understanding what agents actually do: tracing their actions, diagnosing failures, and tracking the cost of execution. He shares author credit on the AgentOps Python package with Shawn Qiu, Braelyn Boynton, and other contributors.[3][5][4][6]

In a coauthored article about AgentOps integration with AutoGen, Reibman and AI engineer Braelyn Boynton argued that production teams need visibility into where agents succeed and fail: “Observability isn’t just an option; it’s a requirement.” They explained that timestamped activity lets developers trace an agent’s actions precisely and understand the reasons behind failures.[6]

Their explanation extended beyond individual model responses to the behavior of the whole system. They described a dashboard that monitors LLM calls, costs, latency, agent failures, multi-agent interactions, and tool usage. Execution replay provides step-by-step graphs of agent activity; their described “Recursive Thought Detection” identifies agents stuck repeating tasks in infinite loops.[6]

The AutoGen integration illustrated an approach to making monitoring easy to adopt: developers supplied an AGENTOPS_API_KEY and called agentops.init() to initialize AgentOps. Reibman and Boynton framed monitoring as part of building and scaling agents, alongside the framework used to create them.[6]

Reibman also teaches agent development. AgentOps lists him alongside Sri Laasya Nutheti as an instructor for Agents 101, covering OpenAI’s Agents SDK, LLM integration, basic agent capabilities, observability, and deployment considerations. At AI Engineer World’s Fair 2025, he served as an Agent Reliability track host alongside PromptQL.[3][1]

The decision to build agent infrastructure grew out of Reibman’s earlier experiments. At a San Francisco hackathon in summer 2023, he tried building web-scraping agents. In an August 2024 TechCrunch interview, he recalled that those agents failed roughly 30–40% of the time, often unexpectedly. His team built internal debugging tools to locate the problems; the tools themselves won the hackathon.[7]

When Reibman demonstrated the tools at subsequent hackathons and events, people asked for access. He described the resulting decision: “Instead of building an agent ourselves, we should build tools to make it easier to build agents.” He started Agency with Adam Silverman and Shawn Qiu, and the debugging tools became its core product, AgentOps.[7]

Before Agency, Reibman worked at consulting firm EY. Agency combined its AgentOps dashboard with consulting to help businesses build agents. In the same 2024 interview, Reibman argued that model providers were unlikely to capture every layer of the agent stack: agent builders would still need surrounding infrastructure to produce production-ready code. That view explained his choice to build tools around agents rather than concentrate solely on an agent application.[7]