Vercel just shipped direct CLI access to anomaly alerts. If you're building on Vercel, you can now pull alert details without touching the dashboard.
The new vercel alerts command lists all alerts for a team or a given project. For each alert, you get the start time, the alert type, and whether it's still active. That's the core loop: know what broke, when it broke, and whether it's still breaking — all from your terminal.
The more interesting part is the AI investigation layer. With the right option, AI investigation results appear alongside each alert. That means when something fires, you're not just getting a signal — you're getting context. The intent is clear: you and your agent can act on alerts without switching contexts.
That last phrase is doing real work. Vercel is explicitly designing this for agentic workflows. An automated agent can now call vercel alerts, read the AI investigation output, and take action — all programmatically. You don't need a human to log into a UI to triage an anomaly.
Observability tooling has historically lived in browsers. Dashboards are fine for humans doing synchronous review. They're bad for agents doing async, continuous monitoring. Pulling structured alert data through a CLI changes the integration surface entirely.
If you're running any kind of automated incident response — whether that's a simple script or a more sophisticated AI agent — this is the primitive you were missing. The AI investigation results being co-located with the alert data in the same command output means your agent gets signal and interpretation in one shot.
If you're running workloads on Vercel, run vercel alerts against your project now just to see what's there. Then think about where this fits in your on-call or incident response flow. If you have any automation that currently polls a dashboard or relies on webhook integrations to catch anomalies, this CLI surface is worth wiring in. And if you're building agents that manage infrastructure or deployments, start prototyping around vercel alerts as an input — the AI investigation output gives your agent something to reason about, not just react to.