Sourcegraph

Today we're launching Agentic Batch Changes in public beta, and we're proud of how it's turning out. Mercari engineer Patrick Klitzke used Agentic Batch Changes during our experimental preview to scope and begin patching a GitHub Actions security vulnerability across their codebase. "I was able to fix it with one prompt on both the Help Center frontend and backend, then extended this to all repos in Mercari. I found around 80 potential repos affected." The thing that made this work wasn't speed. It was the agent's ability to reason about each repository individually. Mercari's repos have similar setups, but not the same setup. A scripted find-and-replace would have collapsed. The agent handled the variation per repo, reacted to CI, and pushed follow-up commits where needed. Canva has been in the preview too, using it to split Batch Changes by code ownership across a Bazel monorepo. This is a public beta. There are rough edges and we're honest about that. But the core capability is working in real customer environments and we're ready to put it in more hands. Available today for Sourcegraph Cloud customers. Free during the beta period. Self-hosted arrives July 8 with Sourcegraph 7.5. Full announcement: sourcegraph.com/agentic-batch-…

Today we launch Agentic Batch Changes into public beta. It is the frontier agent that lets the owner of a massive codebase execute code changes across hundreds (or hundreds of thousands) of repositories at once, starting with a single prompt. It iteratively rolls out the change, self-heals and responds to CI status and PR comments, uses coding agents where judgment is needed and deterministic scripts where not to save you tokens, and provides tracking and auditability to be sure you covered 100% of the places where the update was needed. sourcegraph.com/agentic-batch-…

Heading to AI Engineer World's Fair? Visit Sourcegraph to learn how engineering teams are modernizing large codebases, managing AI-generated change, and building the foundation for reliable coding agents at enterprise scale. Stop by the booth, catch our sessions, and join us after hours. Learn more and schedule time with the team: sourcegraph.com/events/ai-engi…

(2/2) sourcegraph.com/changelog/deep…

Deep Search will now automatically compact long conversations when a follow-up question is close to the context window limit. (1/2)

From our changelog: - Auto-compaction for longer, uninterrupted Deep Search convos - New "Finder" subagent for smarter, token-efficient file discovery - Smart hover summaries are now GA youtube.com/watch?v=yJU01Y…

I’m giving a talk at the @aiDotEngineer World’s Fair 🥳 The talk will be focused on the emerging role of Agent Advocates, what changes when agents are your product users and recommenders, and agent-powered GTM.

You can now have longer Deep Search conversations on Sourcegraph Cloud. We added automatic compaction to help keep your threads manageable as they grow.

Agent failures in large codebases are not random. Data from 1,281 agent runs across 40+ large open source repositories reveals five repeatable failure patterns that show up across coding agents. Each failure pattern points to a different infrastructure problem and a different fix. Read: Why coding agents fail in large codebases (and what to do about it) sourcegraph.com/blog/why-codin…

Pin the dependencies that matter most sourcegraph.com/blog/dependenc…

We automated our security workflow by connecting @Hacker0x01 webhooks to Sourcegraph Deep Search. The system validates and finds fixes for vulnerability reports before you even start work. This helps us address issues faster and removes the manual effort of triaging reports. sourcegraph.com/blog/automatin…

Hope you’re enjoying Microsoft Build! If you use GitHub Copilot, you can connect it to the Sourcegraph MCP server to get context from all your repositories. This works even for your code on GitLab.

Keep track of Deep Search conversations shared with you. We didn't stop there, check out the changelog: sourcegraph.com/changelog/2026…

Gas City maintainer @sgjarmak is running a webinar on building a software factory in enterprise codebases next week. She'll focus on ‘the three asymmetries’ and giving examples using Gas City! Sign up today! sourcegraph.com/webinars/build… #recommended !

See how AI software company @Sourcegraph connects its NetSuite data to Claude to surface financial anomalies faster, ask questions in natural language for real-time answers, and collaborate more effectively. social.ora.cl/6010B8XGfk

.@Sourcegraph analysed 1,281 coding agent runs across large enterprise codebases and found the bottleneck often wasn’t the model — it was context retrieval and navigation. Agents struggled once codebases grew large enough: pulling in the wrong files, thrashing between tools, and making locally correct changes that broke systems elsewhere. One stat says a lot: a task that took one agent 96 tool calls and 84 minutes dropped to five calls and under five minutes with better retrieval infrastructure. The takeaway is this: coding agents need far better context and navigation layers around them to operate reliably in production. 🔗 Read the full story here: tessl.io/blog/coding-ag…

@deepwatrcreatur Cool. Well, we appreciate the video feedback. Anything else you want us to cover?

@Sourcegraph I've long been interested, but time constraints have prevented me from taking a closer look. I hope to do that soon. But these companies offering $20 plans often really want us to use their own clients.

I was watching Sourcegraph in my YouTube feed, remembered that I'm also thinking about code search. Discussed with my agents: are they competitors? Consensus was that semantic search is clearly distinct from filtering for quality, and tracking reasoning behind code decisions.

@deepwatrcreatur nice! So are you a goose user?

@Sourcegraph I just did! It was about agents connecting to your DeepSearch via MCP youtube.com/watch?v=DuJKS0…