The Day Your Codebase Finally Changed Its Own Bed Sheets

If enterprise software had a group chat, the most dreaded message would be: ‘this change needs to happen everywhere.’ It sounds simple. A dependency needs a bump. An API shifts shape. A security hole yawns open like a forgotten basement door. But ‘everywhere’ is where good intentions go to nap. Engineering teams have treated these changes as heroic quests: spreadsheets, side-eye, and someone counting remaining repositories. Sourcegraph is trying to make that saga less tragic with Agentic Batch Changes, now generally available for Sourcegraph Cloud customers. Pricing is based on changesets merged, because impact, not activity, should foot the bill.
I know the feeling. Once I updated a library, strutted away, and called it victory. Ten minutes later, a coworker pinged: ‘Nice. Can you do that in the other seventeen places?’ My hero complex deflated faster than a tire on a nail. That, in miniature, is why scaling code changes is hard. The work is not always complicated. It is just multiplied by a hundred repositories, each with its own little personality, deprecated helpers, and CI gremlins.
Agentic Batch Changes is built to wrangle that zoo. An engineer describes the change once, in plain language. A coordination agent then uses Sourcegraph code intelligence to scope the work across indexed repositories, builds a plan, validates it in a single repository, and pushes out batches of changes. It adapts when repositories differ, reacts when CI fails, and keeps pulling requests forward until they are ready for human review. One prompt, thousands of diffs, one dashboard of merge status. Dan Adler calls the goal simple: change a thousand repositories with single-repo confidence.
The beta already had people testing this on real messes. At Mercari, team lead Patrick Klitzke used the agent to chase a GitHub Actions environment-variable injection vulnerability. One prompt fixed the Help Center frontend and backend, then extended to roughly eighty potential repositories. Unlike a blunt text search-and-replace, the agent could notice similar but not identical setups and handle them with context. At Canva, senior engineer William L. used it for a library migration, raising and merging more than fifty pull requests while tracking status from one interface. No giant catch-all PR, no spreadsheet circus. Just a change that behaved like a tireless intern.
The best part may be the price tag logic. Sourcegraph’s outcome-based pricing charges only when changesets merge. If the team rejects a pull request, it does not count as money changing hands. That clarity matters in an era when AI tools can sound magical but bill like a haunted hotel. Agentic Batch Changes works across GitHub, GitLab, Bitbucket, Azure DevOps, and Gerrit, targeting dependency upgrades, security remediation, code pattern updates, new API rollouts, CI modernization, and end-of-life deadlines. Engineers still keep final approval. No rogue changes, no silent regressions. But for the changes that have always seemed too big, too boring, or too risky to staff, it gives teams something practical: a way to make a codebase-wide fix without summoning the whole department to the emergency room.