Alex
Reibman

At 525 Market Street in San Francisco, Alex Reibman is building the thing every AI engineer eventually realizes they need: a window into what their agents are actually doing. Not what they claim to do. What they do.

AgentOps - the company he co-founded in 2023 with Adam Silverman, Braelyn Boynton, and Shawn Qiu - is an observability platform for AI agents and LLM applications. Think of it as a flight data recorder for AI: session replay, time-travel debugging, real-time cost tracking, audit trails, and comprehensive monitoring across 400+ integrated LLMs and frameworks. Microsoft, Google, Samsung, and Meta use it. So do thousands of dev teams building the next wave of agentic applications.

The origin story is blunt. Reibman started building AI agents that scraped the web. His assessment of how they performed: "They sucked." Rather than iterate in the dark, he decided to build the visibility layer first - the monitoring stack that would let engineers see exactly where and why their agents were failing. That decision became AgentOps.

Reibman came to this by an unusual route. He studied Economics and Philosophy at Emory University - a combination that surfaces later in his writing about the St. Petersburg Paradox and behavioral decision theory. After graduation, he went to Tel Aviv to work as a cybersecurity data scientist at Check Point, where he built classifiers to detect obfuscated malicious JavaScript. It was rigorous, adversarial ML work, the kind where getting the model slightly wrong means something dangerous slips through.

He returned to the US for Ernst & Young, where he led machine learning engineering teams managing engagements worth $40M+ with clients like Goldman Sachs, American Express, and Westpac. That's the world where enterprise AI goes to be taken seriously - and where the gap between a working prototype and a production-grade, auditable system becomes painfully clear.

Every commercial airline flight is required to carry a flight data recorder - a black box that captures every instrument reading, every control input, every system state. The entire purpose is to make failure investigable. AgentOps is building that for AI agents.

The platform's headline feature is session replay: a visual time visualization of every LLM call, tool invocation, action event, and error that occurred during an agent's run. Engineers can rewind, step through frame by frame, and pinpoint exactly where a multi-step workflow diverged from intent. Reibman calls it "time travel debugging."

Underneath the visual layer: real-time cost tracking with live LLM pricing, comprehensive audit trails for compliance, performance optimization tools that can fine-tune specialized models at up to 25x lower cost than frontier models, and enterprise-grade security monitoring. For teams operating at scale, there's SSO, SLA, on-premise deployment, and certifications for SOC-2, HIPAA, and NIST AI RMF.

The market timing is deliberate. By Reibman's research, 89% of surveyed CIOs named agent-based AI a strategic priority for 2025 and beyond. The $5 billion AI agents market is projected to hit $50 billion by 2030. Every one of those deployments will eventually need monitoring. AgentOps intends to be the default choice when they do.

Reibman maintains a blog at alexreibman.com that runs against type for a technical founder. The posts engage with behavioral economics, decision theory, and the mathematics of paradox - the St. Petersburg Paradox, the paradox of choice, privacy as a behavioral problem. The Philosophy degree from Emory surfaces in unexpected places.

He also writes for the Hyper Engineering publication, covering technical deep dives on 100x-ing developer efficiency with AI. On Substack as @hacktrace, he publishes work that sits at the intersection of security thinking and AI engineering.

The through-line in the writing: Reibman is interested in systems that behave in ways their designers don't fully anticipate. That's true for markets with infinite expected value (St. Petersburg), for humans with too many options (paradox of choice), and for AI agents running in production. The observability platform is, in a sense, an engineering answer to a philosophical question: how do you understand a system that surprises you?