Why Did Sourcegraph Replace Cody With Amp Code?

TL;DR

Sourcegraph replaced Cody with the separately branded Amp because its broadly permissioned, tool-calling agent required different workflows, pricing, release cycles, and customer expectations. Amp can ship about 15 times per day without disrupting Cody’s enterprise contracts or cloud and on-premises deployments. Read on to see how that separation supports faster experimentation, growth, and asynchronous agents.

Transcript

Hey everyone, welcome to the latest space podcast. This is Allesio, founder of kernel lips and today there's no Swix. He's in Europe with a engineer, but I'm joined by Quinn and Thirstston from source graph. Welcome. Thanks. Great to be here. Great to be here. So we already had origin of source graph with Byang and Steve. Um so we'll put the the li... Read More

Key Insights

• Amp is a separate product because its tool-calling agent requires different user expectations, pricing, permissions, and development practices from Cody. The separation also protects existing enterprise relationships whose contracts and deployment requirements cannot be changed abruptly.
• A coding agent is defined here as a model combined with a system prompt, tools, and the prompts associated with those tools. Giving the agent broad access to the file system and editor makes its behavior fundamentally different from a retrieval-based assistant or sidebar.
• The central product goal is building the best coding agent, even though the speakers argue that nobody has achieved that goal yet. They describe developer tools as a fast-moving succession in which products can lose their leading position within 6 to 12 months.
• Rapid release capacity is a strategic requirement for Amp because a new model can quickly change what coding agents can do. Separating Amp from Sourcegraph's platform release cycle lets its team ship about 15 times per day and respond faster to technical changes.
• Amp's usage is growing by more than 50 percent month over month, with some weeks growing faster. The speakers also report teams of two or three people reaching annual run rates measured in hundreds of thousands of dollars, showing unusually concentrated use.
• Asynchronous agents could expand coding-agent activity by 10 to 100 times because many agents may run concurrently and continuously in the background. That differs from today's common editor or command-line workflow, where a developer typically works interactively with at most one agent.
• Sourcegraph's established business funds Amp and supplies customer trust, but Amp is not treated simply as another product sold through the same channel. The team accepts inconsistent pricing, limited model choice, and incomplete enterprise checklists to preserve speed while reporting positive gross margins.
• Amp's engineering process began with two experienced engineers working directly on the main branch without code reviews. Ownership and fast repair replaced extensive process, while engineers continually decided whether each situation required lightweight duct-tape execution or more structured, Google-style practices.

Questions & Answers

Q: Why did Sourcegraph replace Cody with Amp Code?

Experiments with Claude models, tools, and broad permissions produced an agent that worked fundamentally differently from Cody. Creating Amp separately let Sourcegraph reset expectations around workflows and pricing while protecting Cody’s enterprise contracts, release cycles, and deployment commitments.

Q: How is Amp different from Cody?

Cody centered on retrieval, chat, assistant panels, and a sidebar, while Amp gives a tool-calling agent broad permission to inspect and interact with the file system or editor. That difference changes how users operate the product and makes Cody’s previous $20 subscription model unsuitable for Amp.

Q: How does Amp define a coding agent?

Amp defines a coding agent as a model combined with a system prompt, tools, and the prompts associated with those tools. The agent receives enough permission to inspect and interact with the file system or editor, distinguishing it from a retrieval-based assistant.

Q: Why does Amp have a separate brand and codebase?

The separate identity frees Amp from expectations attached to Cody and the wider Sourcegraph platform. Its team can change workflows and pricing, move independently of existing contracts, and build quickly enough to respond as models and developer preferences change.

Q: How often does the Amp team ship software?

The Amp team says it has shipped about 15 times per day over the preceding six months. Separating Amp from Sourcegraph’s cloud and on-premises platform release cycles made that pace possible.

Q: How quickly is Amp usage growing?

Amp is reported to be growing by more than 50 percent month over month, with faster growth during some weeks. The speakers also describe teams of two or three people reaching annual run rates measured in hundreds of thousands of dollars.

Q: Why is Amp pursuing asynchronous coding agents?

Asynchronous agents could operate continuously and concurrently in the background instead of requiring a developer to interact with one agent at a time. The speakers expect that shift to enable 10 to 100 times as many agents, increasing both inference use and completed work.

Q: How does Sourcegraph’s established business support Amp?

Sourcegraph’s profitable established product supplies funding and customer trust for Amp’s experimentation. This lets Amp prioritize developers at the model and product frontier without immediately satisfying every security, compliance, legal, pricing, and model-selection expectation of broad enterprise deployment.

Summary & Key Takeaways

• Sourcegraph created Amp after experiments with Claude models and broadly permissioned tools produced a product that behaved differently from Cody. A separate brand, codebase, and operating model allowed the team to reset user expectations, avoid disrupting existing enterprise commitments, and pursue the goal of building the best possible coding agent.

• Amp is organized around rapid adaptation because coding tools and models can change substantially within months. Its team can ship about 15 times per day, sometimes works directly on the main branch without code reviews, and chooses between fast personal-project practices and more structured engineering practices according to the immediate risk.

• Sourcegraph's profitable established product provides revenue and customer trust that support Amp's experimentation. Rather than targeting every developer inside an enterprise, Amp focuses on developers who want to move quickly at the model and product frontier. The team expects concurrent asynchronous agents to increase inference use and completed work by 10 to 100 times.