A note from Dan

If you haven’t read our announcement that Sourcegraph and Amp will become separate companies, go read that first.

‍

I joined Sourcegraph in the fall of 2016. VS Code was just reaching mass adoption as Atom faded; Kubernetes was the hottest new tech that every startup claimed to adopt; and the debate over Angular vs. React was still raging.

Weeks earlier, I was busy shutting down my first startup (a spreadsheet data labeling technology, in fact… maybe I should dust off my patent and see if Meta, Google, or OpenAI are shopping) in the face of a far less interesting pivot.

I found Sourcegraph randomly. I was only a hobbyist coder; I had a degree in CS, but I had gone to consulting, investing, and even business school in the years since. But I clicked on Sourcegraph.com, ran a search, and hopped across a few “go to definitions” and “find references” (which worked perfectly, on the web of all places). My jaw dropped, and I put my hand on my head, and I asked, aloud, “how does every developer in the world not have this?” I wanted to be part of making it happen; I reached out, and the rest is history.

I’ve been here nearly a quarter of my life, from zero ARR to a $2.6bn valuation. I joined Quinn and Beyang years before I even met my wife, and I have a 4 year old now! I sold most of our first contracts, gave our first demos, supported our first customers, built our first dashboards, and my name can be found in git blame across our codebase.

I joined because I was a true believer in the mission and the team. I’m still a true believer in the mission and this team. And I’m ready to build.

‍

Doing it the hard way

For more than a decade, the largest and most forward-thinking software companies in the world have relied on Sourcegraph to understand, navigate, and maintain their code. Long before AI copilots, long before coding agents, long before “context windows” entered the conversation, Sourcegraph was built to solve the hardest problem in large-scale software development:

How do you deeply understand a massive, fast-changing codebase with decades of history?

That question still matters, but for an entirely new reason.

‍

Sourcegraph is the future of code understanding—for humans and agents.

As AI writes code 100x faster, the challenge that humans face onboarding to new codebases, understanding control flows, providing support for undocumented features, and identifying bugs and security vulnerabilities is moving from slow and painful to impossible. Sourcegraph gives you that power back.

Meanwhile, coding agents aren’t omnipotent; they are only as smart as you allow them to be. Access to your code, the system architecture, the semantic graph, the history, and more is essential, and if you, like most developers, work in a codebase that’s too large to fit on your laptop, your agent is blind. Sourcegraph’s enterprise-scale index of your entire codebase is a superpower for agents. 

We believe:

1. Agentic search is the future of code understanding (today, our Deep Search)

2. Every coding agent needs universal code context (today, from our MCP+CLI)

3. The “outer loop” of the software development lifecycle will be automated by AI agents with full code understanding

What AI should be for developers…

We’ve learned a lot of lessons from the creation and growth of Amp, and Deep Search is Sourcegraph’s first bet on agentically searching and aggregating context across vast mono-repos and multi-repo codebases.

We’ve learned the importance of context window management, and have seen how using Deep Search as a sub-agent via MCP can help coding agents like Amp, Claude Code, and Cursor get answers without context window pollution.

We’ve learned that not all human developers (or companies!) are ready to adopt and let agents loose on their codebases. And we’ve learned that agentic search is the perfect bridge: Deep Search has zero barrier to entry, zero user-level setup, zero complexity; but immediate value added. Our users go crazy for this:

“Deep Search gets more unsolicited fan mail than any other dev tool we have ever used” 

“I can already say this is the best thing ever. This is what I imagined AI would do for developers…”

‍

RAG is dead; long live… Search?

But isn't search dead? Or uncool on X at the least? Retrieval augmented generation or RAG, a fancy phrase for using search to provide context to a prompt for an AI assistant, rose and fell so fast it became a meme.

While agents and agentic loops have emerged as the obvious successors to one-shot RAG techniques, those agents still need tools to access that context in a loop. If the agent can’t construct the context it needs, then it will be blind. And it will fail.

Put another way: no new model drop will ever make an agent "good enough" to guess what number you're thinking of. Context and code understanding across massive codebases still matter, agents need visibility beyond a local `grep`. And now, Sourcegraph can power all coding agents.

‍

Agents are coming for the “outer loop”

Change is coming to the “outer loop” of the software development lifecycle, and it will hit us all just as fast and hard as Amp and Claude Code have changed the “inner loop” of code writing. 

Sourcegraph already has roots in key codebase maintenance workflows like large-scale refactoring through Batch Changes, monitoring through Code Monitors, codebase analytics via Code Insights, and more. But they’re complex, and require a ton of investment to get the massive return out. 

What we’re seeing now completely flips that on its head. Agentic refactoring, for example, is an agent that can take a problem statement ("remove all uses of log4j that are affected by CVE-2021-44228”, for example), run deterministic and universal searches across a massive codebase using Sourcegraph, draft prompts and spin up coding agents to fix hundreds or thousands of instances of problematic code, and iterate on that prompt as the sub-agents report back. This is the orchestrator you have dreamed of, one that gives you the peace of mind that it works on all of your code.

The barrier to codebase maintenance will drop to zero. And more is coming.

We are committed to making our API the foundation for our customers and partners to create their own agents, automations, and workflows. We're not working on a code review agent today, for example, but we have customers who are. We will empower you to build the workflows that work for your team with full code understanding.

‍

Join us

Want to work on real, hard problems? Sourcegraph builds cutting-edge agentic tools, fast ranked search, and semantic graphs for customers with a million repositories or multi-terabyte monorepos. We operate at a scale most can’t imagine. This is not Twitter/X hype, this is real, enduring, HARD engineering and customer building. Reach out.