Field note
Vercel put its agent into Slack in public beta for Pro and Enterprise teams. Mentioning @Vercel in a channel, thread or direct message starts a session that reads the discussion and answers with the state of the platform running the app, then turns what the team agreed into changes someone approves. Vercel describes the agent as the first responder for production apps: when something changes it investigates logs, metrics and deployments, proposes a root cause and a fix, and on a pull request flags regressions a passing CI run would not show.
The mechanism is context that is already present rather than context the user supplies. Because the agent is part of the platform, deployments, build statuses, logs, metrics, code reviews and pull requests are in front of it before anyone types, so the expensive part of an incident conversation — assembling the evidence — is skipped rather than delegated.
That is the distinction drawn in an organization agent being a scoped runtime rather than one Slack bot: the chat surface is where intent is expressed, but the authority and the evidence live in the platform behind it. The agents hub tracks that split as more vendors put a mention handle in front of production systems.
This is a public beta announcement, and the approval step is described rather than specified. The condition to watch is whether a Slack-initiated change carries the same review record as one opened from a terminal, because an agreement in a thread is not an audit trail.