Sourcegraph makes Agentic Batch Changes generally available for Cloud customers

Sourcegraph said Agentic Batch Changes is generally available for Sourcegraph Cloud customers — an AI agent that plans, runs, and tracks a code change across hundreds or thousands of repositories from one prompt, with pricing based on the pull requests that actually merge.

Coding agents already write in one repo. The hard job is the same upgrade or security fix across a whole company codebase. Sourcegraph’s same-day primary pairs a fleet-wide change agent with pay-for-what-merges pricing — treat the beta merge counts as company figures.

Sourcegraph, a San Francisco code-intelligence company, said via Business Wire on 14 Sep 2026 that Agentic Batch Changes is generally available for Sourcegraph Cloud customers. Code intelligence here means a searchable index of a company’s code so people and software can find and change it. A repository, or repo, is one project’s code folder. Cloud means Sourcegraph’s hosted product, not a laptop install. That company release is the filing event. These are company claims.

A batch change, as Sourcegraph uses it, is one planned edit that has to land in many repos — a shared-library upgrade, an API fix, a security patch. The older Batch Changes product already opened pull requests — proposed code edits for review — across those repos and tracked them until they merged. A changeset is Sourcegraph’s word for one of those proposed edits. CEO Dan Adler, quoted on the release, said Batch Changes already merges half a million changesets a year “for the largest enterprises in the world,” and that Agentic Batch Changes “has already merged nearly a thousand changesets for early adopters during our beta.”.

Adler’s problem statement, as the same page has it: every engineering leader has a list of changes that need to happen everywhere, and “those are the projects that never get staffed.” The release says a single migration can turn into hundreds of pull requests, and that nobody volunteers to upgrade a framework across 10,000 repositories.

With Agentic Batch Changes, the company said, an engineer describes the change once in plain language. A coordination agent — software that plans the job and assigns the work — uses Sourcegraph’s index to find the affected repos, builds a plan, and tries the approach in a single repository before generating the full set of diffs. A diff is the line-by-line edit. It then pushes changes in batches, handles repo-to-repo differences, reacts when CI fails — CI, or continuous integration, is the automatic test run that starts when code is proposed — and keeps going until the pull requests are ready for an engineer to review and merge.

Engineers review and approve every changeset before it merges, the company said. This is not a claim that the agent merges on its own.

With general availability, Sourcegraph introduces outcome-based pricing: the coordination agent is priced on the number of changesets merged into the customer’s codebase. If it opens a pull request and the team does not merge it, the customer does not pay for that pull request.

During the beta, the company said, customers merged nearly a thousand changesets on work that ranged from security fixes to library migrations. Mercari, a Japan-based marketplace, is named. Patrick Klitzke, a team lead there, said he used one prompt to fix a GitHub Actions environment-variable injection issue — a flaw where a workflow can leak or run the wrong secrets if environment variables are set badly — on the Help Center frontend and backend, then extended the search across Mercari and found around 80 potential repos affected. Quoted: with Agentic Batch Changes, “you're able to handle repos that have similar, but not identical setups.”.

At Canva, the named job was a library migration. William L., a senior software engineer there, said the team used Agentic Batch Changes to raise and merge 50+ pull requests and track each one from one web screen. Quoted: without it, “we would have been managing either huge PRs or spreadsheets.” File 50+ and the quote as his, via Sourcegraph.

Agentic Batch Changes is generally available as of 14 Sep 2026, the release says. Named code hosts: GitHub, GitLab, Bitbucket Server, Bitbucket Cloud, Azure DevOps, and Gerrit.

For each part of a plan, the company said, the coordination agent decides whether the change needs a coding agent or, more often, a script, so it does not pay a coding agent to repeat a mechanical edit a hundred times. Where a change needs judgment, it can hand work to Claude Code or Codex — two named coding agents — with repo-specific instructions and codebase context through Sourcegraph MCP. MCP, or Model Context Protocol, is a standard way for an AI tool to read extra context from another system. When CI fails, the agent adjusts and tries again. How those agents are wired stays in Sources.