About the job
At AMD, we are redefining the paradigm of low-level system software development. We are seeking a Senior Staff Software Engineer to develop the deep integration of high-autonomy agents (e.g., Claude Code, Cursor) into our system software development workflow. In this role, you will architect an automated, closed-loop system - from requirement description to engineering task verification. By leveraging Verification-Driven Engineering and Feasibility Analysis to determine solvability within existing frameworks, while enabling the Agent to autonomously acquire and synthesize new skills through iterative self-learning, you will transform complex low-level engineering challenges into predictable, controlled agentic tasks, while architecting autonomous agents capable of independent problem-solving and self-evolving technical skills.
Responsibilities
1. Architect Verification-Driven Agentic Workflows
Multi-Agent Collaboration: Implement strategies involving specialized roles (e.g., infra-Architect, Debug-Coder, QA-Validator) to ensure high-quality engineering task output and minimize hallucinations.
Domain-Knowledge Centric RAG: Build high-precision retrieval systems using LangChain (LCEL) to index massive repositories, PDFs, and Confluence pages, utilizing advanced strategies like Parent Document Retrieval and Semantic Chunking.
Complex State Machines: Design and implement cyclic, multi-step reasoning architectures using LangGraph to manage long-running coding tasks and "reflection" loops.
2. Autonomous Execution & Self-Correction
Zero-Touch Provisioning: Develop systems where agents autonomously set up sandboxed runtimes, resolve dependencies, and configure infrastructure.
Autonomous Test Synthesis: Architect engines that generate edge-case reproduction scripts and validate fixes within isolated CI/CD pipelines.
Self-Healing Remediation: Engineer loops that enable agents to parse execution logs, identify root causes, and iteratively apply patches until tests pass.
3. Benchmarking & Optimization
Performance Evaluation: Lead the evaluation of agentic performance using industry-standard benchmarks (e.g., SWE-bench), aiming for top-tier recovery rates.
Trace Analysis: Utilize LangSmith for deep trace analysis, debugging complex agent trajectories, and optimizing prompt/chain latency and cost.
Qualifications
Minimum
No minimum qualifications listed.
Preferred
AI Agent Architecture: Proficient in architecting autonomous AI agents using LangGraph, AutoGen, and LangChain. Proven experience in building self-correcting engineering workflows and validating performance via benchmarks like SWE-bench.
System Programming Excellence: Deep experience in C/C++, with expert knowledge of Linux, memory management, and interrupt handling. Familiar with modern software development process, including complex CI/CD pipelines.
MCP and Skills Development: Experience in custom MCP Servers and Skills.
AI Developer Insight: Advanced user of AI tools (Cursor, Claude Code) or developer of LLM-based agentic plugins. Deep understanding of Prompt Engineering and debugging strategies for non-deterministic systems.
Engineering Philosophy: Strong belief in "Verification as the Boundary." Ability to decompose complex NP-level engineering problems into automatically verifiable P-level tasks.