Engineering Teams Restructure Around AI Agents as Code Review Becomes New Bottleneck
San Francisco, CA — Last night, at least 10 events across the city aimed at connecting AI startups with venture capitalists highlighted a quiet but profound shift: engineering teams are reorganizing around AI agents. The most notable gathering, Camp AI's 'Agents at Work' event hosted by Auth0, showcased companies already deep into this transformation.
Companies like Browserbase, Mastra, Fireworks AI, Drata, Mya, MindFort, and Corridor are building the vendor ecosystem to enable secure, performant agentic AI. But the real revelations came from their own internal restructurings — and the new bottlenecks emerging.
Smaller Teams, Broader Scope
Paul Klein IV, founder and CEO of Browserbase, delivered the night’s most striking observation: 'If AI is not doing your whole job it’s a skill issue at this point.' His comment underscored the speed at which AI adoption is reshaping engineering roles.
Abhi Aiyer, founder and CTO of Mastra, explained that teams are shrinking even as their output expands. 'You can have one person run a whole feature project because they have an army of one to infinity AI agents behind them,' Aiyer said.
The Code Review Crisis
Multiple panelists argued that AI-generated code has moved the bottleneck from writing software to safely reviewing and operationalizing it. Aiyer noted that engineering teams are opening significantly more pull requests, while review throughput struggles to keep pace.
Klein stressed the need for throttling experimental AI output. 'If you are in the critical path and customer facing, no slop. If you are not critical path, not customer facing, slop away,' he said.
Trust, Ownership, and Observability
Speakers repeatedly flagged ownership as a key challenge. Rob Ferguson, VP of technology and strategy at Fireworks AI, asserted that responsibility cannot vanish simply because code was generated by AI. 'It doesn’t matter if you typed it or prompted it, you own it,' Ferguson said.
Bhavin Shah, VP of AI product at Drata, added that enterprise systems now require constant 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 demonstrations focused 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 specifically to secure agent interactions.
Monica Bajaj, SVP of engineering at Okta, emphasized minimizing risk exposure as agents operate autonomously across enterprise systems. 'How do we make sure that those tokens are not long-lived tokens?' she said.
Background
The event was part of a broader trend: AI startups are rapidly adopting agentic architectures. The vendor ecosystem includes companies providing infrastructure for secure, performant agents, but the shift is forcing engineering teams to redesign workflows, review processes, and security models.
Key players like Browserbase, Mastra, and Fireworks AI are experiencing the transition internally as they build tools for others. Their firsthand stories reveal both the potential and the friction points.
What This Means
Engineering teams will continue to shrink in size but expand in scope, with AI agents handling more of the development lifecycle. However, the bottleneck will shift to code review, security, and governance. Organizations must invest in automated review tools, throttling mechanisms, and robust authentication to keep pace.
Ownership and accountability remain human responsibilities, even when code is AI-generated. Observability and audit trails will become mandatory for enterprise adoption. The competitive advantage will belong to teams that can safely scale agentic workflows without sacrificing quality or security.