Product Manager, AI Platform SW - Agentic AI Kernel Generation

Description

The company's AI software platforms team is building the next generation of agentic AI infrastructure, where coding agents write, tune and ship GPU kernels with little manual work.

The team builds AI kernels that link data pipelines, evaluation suites and GPU-accelerated runtimes, so developers can release inference and training software sooner and with better performance.

As the product manager, you will be the internal advocate for AI agents and LLM-based coding workflows that generate optimized kernels. You will work with engineering, research and customers to set strategy, build roadmaps and create products covering the full agent lifecycle: data collection, synthetic data generation, evaluation, deployment and ongoing improvement.

This is an existing vacancy.

Responsibilities

• Design agent-focused products that let coding agents generate, refactor and optimize CUDA kernels and graph-level execution plans for several GPU architectures.
• Own the full data lifecycle for agent training and evaluation, covering dataset curation, synthetic data creation and benchmark suites that measure correctness, latency and adaptability.
• Work with CUDA, kernel and compiler engineering teams to connect agents to compilers, profilers, execution sandboxes and runtimes in a way that is safe and observable.
• Work with internal and external developers, company leaders and ecosystem partners to coordinate multi-agent orchestration, set feature priorities, and plan launches and their messaging for agentic AI kernel generation.

Requirements

• Have 7+ years of technical product management or closely related experience shipping developer or platform products in AI, ML infrastructure or high-performance computing, with a strong record of end-to-end ownership and impact.
• Have proven experience in the AI agent or LLM space, including building or productizing coding agents.
• Have experience with multi-agent orchestration and self-healing or code loops that improve over time (required).
• Have worked on connecting agents to compilers or execution environments.
• Have a proven record of creating and releasing automated testing or evaluation suites that measure agents on objective metrics such as correctness, performance and latency, and use the results to guide development.
• Hold a BS or MS in Computer Engineering, Computer Science or a related technical field, or have equivalent experience in parallel computing architectures and systems.
• Ways to stand out from the crowd: PhD or equivalent experience in Computer Engineering, Computer Science or another technical specialty.
• Ways to stand out from the crowd: Experience building or launching coding-agent platforms or copilots used by development teams at scale.
• Ways to stand out from the crowd: Contributions to performance-critical open-source projects, such as Triton, TVM, FlashAttention, kernel libraries or agent frameworks, with clear community adoption and impact.
• Ways to stand out from the crowd: Research in GPU kernel optimization, collective or group communication algorithms, multi-agent systems, or ML model serving and inference architecture, showing how you reason about systems end to end.
• Ways to stand out from the crowd: Experience building cost-per-inference or cost-per-token models that account for hardware utilization, energy efficiency and cluster scaling, and using them to guide product strategy and tradeoffs.

Conditions

Base pay depends on location, experience and other objective job-related factors.

Level 4 base salary range: 168,000 USD to 258,750 USD.

Level 5 base salary range: 208,000 USD to 327,750 USD.

Applications will be accepted at least until November 3, 2026.

The company uses AI tools in its recruiting process.

The company is an equal opportunity employer and does not discriminate on the basis of race, religion, color, national origin, gender, gender expression, sexual orientation, age, marital status, veteran status, disability status or any other legally protected characteristic.