Vercel Agent enters public beta with AI code review and production debugging
Vercel has launched Vercel Agent, an AI-powered teammate designed to assist with development workflows. Now available in Public Beta, the agent combines AI with platform expertise, application code, and telemetry data from Vercel to help teams ship faster while maintaining quality. The initial release includes two core skills: Code Review and Investigations.
Vercel Agent is positioned as a multi-purpose tool whose capabilities will expand over time. Its key differentiator is validation: every suggestion or investigation is grounded in the actual state of the project’s code, framework, and infrastructure, then verified end-to-end before being presented. This means developers receive tested, actionable recommendations rather than unverified guesses.
Code review with sandbox verification
The Code Review skill works within GitHub workflows, reviewing pull requests with awareness of the entire project stack. Rather than just analyzing diffs statically, the agent orchestrates a multi-step reasoning process that examines dependencies and framework patterns across the codebase. It references lint and test results, runs targeted verifications, and generates feedback that passes framework-level checks automatically.
The practical effect is faster merges without sacrificing quality. The agent can identify regressions before they reach production, flag performance issues that static analysis might miss, and suggest fixes that are both secure and ready for production. For developers, this translates to having a reviewer that understands the full context of the application and tests every recommendation in a simulated build environment inside a Vercel Sandbox.
Investigations for production issues
The Investigations skill extends the agent’s reasoning capabilities to the runtime. When unusual application activity occurs, the agent analyzes both real-time and historical data to determine why the event happened. Using Vercel’s Framework-defined infrastructure, it has deep context about the application, allowing it to interpret underlying data and identify the cause of spikes, slowdowns, failures, or unexpected changes.
Investigations can answer questions such as why a deployment slowed down, which function is throwing errors in production, or what changed between two builds. Each investigation is based on the actual state of the application, not speculation. For engineering teams, this turns hours of debugging into seconds of reasoning, converting observability from complex dashboards into a straightforward conversation about system health.
Agent-as-a-Service and what’s next
Vercel describes Vercel Agent as the first product in a new category they call Agent-as-a-Service. Similar to how Software-as-a-Service abstracted infrastructure complexity, Agent-as-a-Service means teams can rely on a production-grade agent that runs securely on their behalf without building or hosting it themselves. A new agentic marketplace is also available to help users get started.
With Code Review and Investigations now live, Vercel plans to expand the agent’s capabilities beyond diagnosis. Upcoming features will allow the agent to create pull requests, propose fixes, and automate workflows from detection through to deployment. The agent runs on the same infrastructure that powers Vercel itself, built to support the confidence to move fast without cutting corners.
