From Cody to Amp: Our Journey into Agentic Programming

Our team has always been at the forefront of adopting tools that enhance our development workflow. For a long time, Sourcegraph Cody was our go-to AI assistant. It provided valuable code suggestions and context-aware completions that streamlined our daily tasks. However, as our projects grew in complexity, we began to encounter limitations that prompted us to explore alternatives.

Enter Sourcegraph Amp—a revolutionary agentic coding tool that has transformed the way we approach software development.

The Limitations of Cody

While Cody served us well, we faced several challenges:

• Manual Code Acceptance: Cody required us to manually accept code changes, which disrupted our flow and added overhead.

• Context Limitations: The tool often ran out of context or token space during complex tasks, forcing us to refactor and re-prompt.

• Lack of Multi-Step Awareness: Cody lacked the ability to understand and manage multi-step tasks or maintain awareness of to-do lists.

• Limited Long-Running Operations: It struggled with long-running operations, often failing to complete large-scale tasks.

These limitations led us to seek a more robust solution.

Discovering Amp: A Game-Changer

After just four hours with Amp, we knew we had found our new development companion. We promptly transitioned from Cody and invested in Amp, recognizing its immense value.

Key Features That Won Us Over

• Automatic Code Changes: Amp applies code changes directly to source files without requiring manual acceptance, streamlining our workflow.

• Enhanced Context Management: With the integration of Model Context Protocol (MCP) servers, Amp provides better automatic context, allowing for more informed code suggestions.

• Agentic Awareness: Amp exhibits multi-step agentic awareness, effectively managing to-do lists and actions, ensuring tasks are completed comprehensively.

• Sub-Agents for Context Retrieval: It employs sub-agents that autonomously retrieve documentation or additional context, enhancing its decision-making capabilities.

• Agent.MD Integration: Amp's agents access Agent.MD to load pertinent context into each coding session interaction, ensuring that every action is informed by the most relevant information.

• Support for Long-Running Operations: Amp handles longer operations seamlessly. We recently left it running on a large task for 20 minutes, resulting in the generation of approximately 6,000 lines of tests.

• Integration with Git Issues: The tool connects with Git issues, enabling semi-automatic completion and better project management.

• Standalone CLI: Beyond the IDE, Amp offers a standalone Command-Line Interface (CLI), expanding its versatility and integration into various workflows.

MCP Features and PostgreSQL Integration

Amp's MCP features have dazzled us as we integrate with both Context7 and PostgreSQL, further enabling Amp to assist in the software development process. The PostgreSQL MCP server provides read-only access to databases, allowing Amp to inspect schemas and execute queries, enhancing its understanding of our data structures.

• Context7 for Up-to-Date Documentation: By integrating with Context7, Amp fetches real-time, version-specific documentation and code examples directly from the source, eliminating outdated or hallucinated code suggestions.

Autonomous Build Execution and Error Handling

Amp's ability to run builds autonomously and take corrective actions when builds fail has greatly improved our output. This feature ensures that issues are identified and resolved promptly, reducing downtime and enhancing productivity.

The Impact on Our Team

Transitioning to Amp has significantly accelerated our development velocity. The tool's autonomous reasoning and comprehensive code editing capabilities have reduced manual overhead and allowed us to focus on more strategic aspects of our projects.

We're particularly excited about the possibilities Amp opens up beyond the IDE. Its standalone CLI means we can integrate it into our broader development processes, further enhancing our productivity.

Conclusion

Our journey from Cody to Amp has been transformative. While Cody was a valuable tool, Amp has redefined our expectations for AI-assisted programming. It's not just an assistant; it's a collaborative partner that understands our projects, anticipates our needs, and actively contributes to our success.

For teams seeking to elevate their development workflow, we wholeheartedly recommend exploring Sourcegraph Amp. It's a testament to what AI-assisted programming can and should be.