In San Francisco, where innovation pulses through every startup hub and coffee-fueled late-night session, AI hackathons are quickly becoming the city’s real weekend scene. While others unwind at the beach or in the city’s vibrant nightlife, the Bay Area’s most ambitious minds are building the future — line by line, prompt by prompt.
One such future-shaping project is Agency, a startup born not from a VC pitch deck but from the floors of San Francisco’s AI hackathons. What began as an experimental tool for debugging malfunctioning AI agents has now evolved into a crucial infrastructure layer for anyone building with autonomous AI systems.
A Problem, a Solution, and a Pivot
It all started last summer, when Agency AI co-founder Alex Reibman set out to build web-scraping AI agents — bots designed to perform complex tasks using public internet data. But the results were, in his words, disappointing.
“The agents failed like 30 to 40% of the time, and often in unexpected ways,” Reibman recalls
To make the agents more reliable, his team built internal tools to track exactly what each agent was doing and where things were going wrong. These debugging tools turned out to be more useful — and more impressive — than the agents themselves, eventually winning the hackathon.
That was the spark.
“People started asking for access to the tools. That’s when I realized: the real opportunity isn’t building the agents — it’s empowering others to do it better,” says Reibman
Introducing Agency: AgentOps for the New AI Era
Reibman teamed up with Adam Silverman and Shawn Qiu to officially launch Agency, a company focused on solving one of the biggest headaches in AI development: observability and control for autonomous agents.
Their flagship product, AgentOps, is a dashboard and infrastructure platform designed to help developers understand, monitor, and debug AI agents in real time. From showing step-by-step task execution to cost tracking and behavior prediction, it’s like a mission control system for your bots.
“Think of AgentOps as multi-device management, but for AI agents,” explains Silverman. “You can visualize the guardrails you’ve set — and whether your agents are respecting them — before you move into production”
Working Across the Ecosystem
Agency isn’t trying to compete with the major players like OpenAI, Anthropic, or Google. Instead, it complements their offerings by sitting within the growing AI stack — a necessary tool in making AI agents production-ready.
It’s also model-agnostic, working with frameworks like AutoGen, crewAI, and AutoGPT, and integrating with AI model providers like Cohere and Mistral. This flexibility makes Agency ideal for teams working across various agent systems and languages.
From Blog Bots to Hedge Funds
While Agency keeps its customer list private, Reibman shares that its tools are already in use by hedge funds, consulting firms, and marketing agencies. One client even built an AI agent that writes blog posts for portfolio companies — and now uses AgentOps to track performance, flag issues, and control costs.
The startup recently raised $2.6 million in pre-seed funding, led by 645 Ventures and Afore Capital, a strong sign of investor confidence in the infrastructure side of the AI agent wave.
What’s Next for AI Agents — and for Agency
With the agent ecosystem still in its infancy, and big names poised to enter the field in force, Agency is playing a smart long game: help others build better, safer, and more efficient agents, without trying to own the entire stack.
“OpenAI might build the agent builders, but there are so many other layers needed to make sure the code is production-ready,” says Reibman
In a landscape where AI agents are starting to act more independently, transparency and control are everything — and that’s exactly what Agency offers. Born out of late nights, failed demos, and unexpected insights, it’s a startup that reflects the true spirit of San Francisco’s hackathon culture: build fast, fail forward, and fix the real problem.
