Senior Software Engineer - Autonomous Coding Agents

About Lexsi Labs

Lexsi Labs is a cutting-edge AI laboratory dedicated to creating aligned, interpretable, and safe superintelligent systems. Our immediate goal focuses on developing safety-conscious autonomous systems operational in the near future. Our research covers AI alignment methods, interpretability-driven system architectures, and foundational model studies in structured data and autonomous system design. Over the last 15 months, we have contributed over 25 papers to top conferences such as ICLR, ICML, WWW, IJCNN, MICCAI, and Eurips. We maintain labs across Mumbai, Paris, and London and foster a flat organizational structure where engineers enjoy high autonomy and ownership from design to production.

Role Overview

In our current development phase, we are creating autonomous systems to tackle complex challenges in software engineering, data science, and AI research. This includes building infrastructure components intended to operate within customers’ environments rather than ours. As a Senior Software Engineer, you will be responsible for the coding agent—a system that executes objectives by interacting with code repositories, autonomously verifies its changes, and systematically documents its actions. This agent works with large, legacy, lightly tested, critical codebases within customer networks, often functioning without internet access.

Key Responsibilities

• Design and enhance the agent architecture and its tooling capabilities, including code retrieval, editing, building, testing, static analysis, and version control integration.
• Implement verification mechanisms to ensure changes are validated as correct before human review.
• Develop long-term execution strategies to handle multi-file refactoring and migrations with support for checkpointing, recovery, and self-correction.
• Define authorization and rollback protocols that regulate agent operations and their reversibility.
• Integrate the agent within customers’ engineering environments and develop observability tools to facilitate debugging and auditing.
• Create evaluation suites based on real-world repository tasks and fine-tune agent behavior based on these assessments.

Collaboration

Work closely with our research teams on designing evaluation systems, post-training analysis, and improving interpretability of the agent's behavior.

Candidate Profile

• Extensive expertise in developer tooling and integrated development environments (IDEs), including editor internals, extensions, language servers (LSP), code intelligence, and refactoring engines, with a track record of delivering tools widely used by engineers.
• Strong background in code intelligence and program analysis like ASTs, tree-sitter, static and dataflow analyses, symbol indexing, code search, call graph construction, dependency resolution, codemods, and automated migration tooling.
• Experience with build systems and continuous integration/delivery (CI/CD) including Bazel, Gradle, Maven, Make, incremental and hermetic builds, artifact and dependency management, and pipeline design involving Jenkins, GitLab, GitHub Actions, including offline and self-hosted deployment scenarios.
• Proficiency in observability technologies such as distributed tracing, structured logging, OpenTelemetry, and developing tooling for replay and debugging of non-deterministic systems.
• Skilled in integration engineering across Git hosting APIs, ticketing, workflow systems, enterprise authentication, and working with legacy self-hosted software versions.
• Experience implementing agentic systems in production environments with awareness of current agent frameworks like ReAct agents, LangGraph, LangChain, and Semantic Kernel along with their limitations.
• Strong backend engineering fundamentals in advanced Python development, API and service design, sandboxing, containerization, database and data pipeline management, and knowledge of cloud as well as on-premise infrastructures.
• Ability to balance multiple constraints including performance, reliability, cost, safety, and interpretability holistically.
• Comfort with ambiguity, making good judgments unprompted, and taking ownership where specifications are loose.

Additional Advantages

• Experience in building evaluation datasets or frameworks specifically for AI systems.
• Background in compiler technologies, managing large monorepos, or leading legacy system modernization projects.
• Exposure to production environments focused on alignment, interpretability, or safety tooling.

Working Culture

Our organization moves at a rapid pace and we expect candidates to keep up accordingly. We prioritize substantive work and effective execution over superficial presentation or rhetoric.