I counted at least 10 events in San Francisco last night aimed at matching AI startups with VCs. Just another Thursday.
But what made Camp AI’s “Agents at Work” event (hosted by Auth0) stand out was its showcase of companies that are in various stages of reorganizing their engineering processes around AI agents. Browserbase, Mastra, Fireworks AI, Drata, Mya, MindFort, and Corridor are all part of the vendor ecosystem trying to enable secure and performant agentic AI, but the most revelatory stories were their own successes and the challenges they faced restructuring their engineering orgs for agents.
Agentic AI is reshaping team structures
Paul Klein IV, founder and CEO of Browserbase, delivered the night’s most memorable line while discussing the speed of AI adoption inside engineering teams. “If AI is not doing your whole job it’s a skill issue at this point,” said Klein.
Abhi Aiyer, founder and CTO of Mastra, said the result is dramatically smaller teams capable of executing much larger scopes of work. “You can have one person run a whole feature project because they have an army of one to infinity AI agents behind them,” said Aiyer.
The AI-generated code bottlenecks have moved
Several panelists argued that AI systems are now generating software faster than organizations can safely review and operationalize it. Aiyer said that engineering teams are opening significantly more pull requests while review throughput becomes the new bottleneck.
Klein stressed the importance of throttling experimental AI output to appropriately lower risk in deployment environments. “If you are in the critical path and customer facing, no slop,” he said. “If you are not critical path, not customer facing, slop away.”
Trust and ownership are common stumbling blocks
Speakers repeatedly emphasized observability and accountability as challenge areas for autonomous agents. Rob Ferguson, VP of technology and strategy at Fireworks AI, argued that ownership cannot disappear simply because AI generated the output. “It doesn’t matter if you typed it or prompted it, you own it,” Ferguson said.
Bhavin Shah, VP of AI product at Drata, said enterprise AI systems increasingly require detailed auditability. “The agent is constantly telling the user, here is the action I’m taking, here is what I’ve done,” he said.
Securing the agentic workflow
Auth0’s demos focused heavily on authentication, authorization, and runtime controls for AI agents interacting with APIs and Model Context Protocol (MCP) servers. The company’s new MCP authentication product, which reached general availability this week, is designed to secure how agents interact with MCP servers and APIs.
Monica Bajaj, SVP of engineering at Okta, emphasized the importance of minimizing risk exposure as agents operate autonomously across enterprise systems. “How do we make sure that those tokens are not long-lived tokens?” she asked, adding “We make sure that the blast radius is minimum.”
Infrastructure is becoming the real AI differentiator
Klein argued that many AI limitations today are no longer about the underlying models themselves. “The overhang of AI capabilities is actually an infrastructure problem, not a model quality problem,” he said.
Klein noted that orchestration, tooling, permissions, and training data pipelines increasingly determine whether AI systems succeed in production.
Other interesting company demos
Mya demonstrated an AI program manager that aggregates Slack, Gmail, Jira, GitHub, and meeting notes to automatically track project risk and operational status. MindFort showed autonomous penetration testing agents designed to continuously probe enterprise applications for vulnerabilities during development and runtime. And Corridor demonstrated AI security guardrails that pre index codebases and inject secure coding guidance directly into AI coding workflows.
Mastra discussed redesigning developer documentation and frameworks specifically for AI agents rather than human developers.