Agent Engineer [IC4]

About the role

Sourcegraph is dedicated to enhancing the way engineering teams manage complex codebases. As AI continues to transform code generation, we aim to provide the necessary infrastructure for understanding and evolving that code. Our tools, including Code Search, Deep Search, MCP, and Agentic Batch Changes, empower teams to navigate and modify extensive codebases effectively. If you're interested in shaping the future of engineering practices, we invite you to join our team. You will own the design and implementation of agentic systems that directly enhance user experiences across our core product offerings. You will be responsible for establishing robust evaluation strategies to measure the real-world effectiveness of changes in our agentic products. A key part of your ownership involves selecting, upgrading, and training models to optimize their performance specifically for our applications and use cases. You will engineer critical retrieval and context mechanisms to improve the accuracy and reliability of model outputs. Managing cost and latency considerations will be essential to ensure sustainable and scalable product development. You will mentor peers and contribute to team knowledge in agent engineering practices to elevate the overall capability of the Code Understanding team.

Key facts

What you'll do

• Lead the development of agentic systems that enhance user experiences across our product offerings.
• Implement evaluation strategies to assess the effectiveness of changes in our agentic products.
• Select, upgrade, and train models to optimize their performance for our applications.
• Engineer retrieval and context mechanisms to improve the accuracy of model outputs.
• Manage cost and latency considerations to ensure sustainable product development.
• Architect and maintain the infrastructure required to support complex agent workflows.
• Analyze user interaction data to identify opportunities for improving agent reliability and usability.
• Collaborate closely with product managers to define success metrics for new agentic features.
• Develop and maintain testing suites that validate the behavior of agentic workflows.
• Translate high-level product goals into technical specifications for agentic functionality.
• Investigate and resolve complex technical issues that arise in production agentic systems.
• Partner with frontend teams to ensure seamless integration of agentic capabilities into the user interface.
• Document architectural decisions and operational procedures for long-term maintainability.
• Mentor junior engineers on best practices for building reliable and effective agentic systems.

Requirements

• Proven experience in agent engineering, combining software development, machine learning, and statistical analysis.
• Ability to navigate and solve complex problems independently, with a focus on high-risk projects.
• Strong background in software engineering, including proficiency in Go, TypeScript, GraphQL, Postgres, and Docker.
• Familiarity with agentic coding tools and the ability to work effectively in a fast-paced, remote environment.
• Experience in mentoring peers and contributing to team knowledge in agent engineering practices.
• Demonstrated capability to manage multiple priorities in a dynamic and evolving work setting.
• A track record of delivering production-grade software that scales to meet user demands.
• Strong written communication skills for effective collaboration in an asynchronous work environment.

Nice to have

• Experience in launching LLM-powered or agent-based developer products.
• Background in building evaluation frameworks for machine learning products.
• Knowledge of retrieval, ranking, and search relevance techniques.
• Direct experience with enterprise clients and translating their requirements into actionable product features.

Skills & tools

• Proficient in software engineering with a strong focus on production services.
• Comfortable working across our technology stack, including backend and frontend technologies.
• Understanding of agentic coding tools and the implications of the code they generate.
• Adaptable to an asynchronous, multi-service work environment.

Practical notes

This position is open to candidates globally, with a preference for those located in Europe or North America. Applicants should be available for at least 20 hours of overlap with Eastern Standard Time.