Dependency updates with an agent instead of Renovate bots

Renovate opens the PR and stops. An agent reads the changelog, fixes the breaking changes, and hands you a green merge request.

Renovate is great at noticing that a new version exists. It is not great at making your code work with it. That gap is where a coding agent earns its keep.

Where bots stop

A bot-driven update is a version bump and a lockfile change. For patch releases that is usually enough. For major and feature releases it rarely is:

The result is a familiar backlog: a wall of red Renovate PRs that everyone scrolls past. The updates that matter most, the majors, are exactly the ones that stay unmerged the longest.

What an agent does differently

An agent treats the update as a task, not a bump:

  1. Reads the changelog and migration guide for the new version.
  2. Applies the version bump, then fixes the call sites the release actually broke: renamed parameters, moved imports, changed config keys.
  3. Runs your test suite and iterates until CI is green.
  4. Opens a merge request with the reasoning in the description, ready for human review.

You still review and you still merge. The difference is what lands on your desk: a green MR with the migration already done, instead of a red diff and a link to release notes.

Running this on vemlor

vemlor is the control plane that makes this a workflow instead of a science project. A failing update PR or a webhook becomes a task. A runner executes your agent, Claude Code or Codex on your own subscription, against the affected repo. Every step is logged, and the output is a reviewed merge request on GitHub or GitLab.

You keep Renovate or any other bot as the detector if you like. The agent takes over where the bot stops.

Next steps