Product Manager, Agent Harness & Modelling

Who are we?

Cohere is the leading security-first enterprise AI company. We build cutting-edge foundation AI models and end-to-end products that are designed to solve real-world business problems.

We’re training and deploying frontier models for enterprises who are building AI systems. We believe that our work is instrumental to the widespread adoption of AI and we are looking for folks that want to be part of that.

We obsess over what we build. Each one of us is responsible for contributing to increasing the capabilities of our models and the value they drive for our customers. Cohere is a team of researchers, engineers, designers, and more, who are all passionate about their craft.

We are a global technology company co-headquartered in Toronto and San Francisco, with key offices in London, New York City, Montreal, Seoul, Germany and Paris. Join us!

Cohere is revolutionizing enterprise AI with North, an agentic AI platform designed to securely deploy AI agents and automations within organizations' infrastructure. North empowers employees to streamline workflows, automate repetitive tasks, and unlock actionable insights while ensuring data privacy and compliance. North combines cutting-edge generative and search models with customizable integrations to drive productivity and innovation at scale.

We are seeking an Agent Harness Product Manager to own the execution layer that makes North agents reliable, capable, and production-ready. This is a role that sits at the intersection of three domains:

• Agent Loop and Execution:
  Own the core agent runtime: tool orchestration, parallel execution, sub-agent delegation, sandbox code execution, and failure recovery. You will define how North agents plan and act across long, multi-step workflows and ensure the execution environment is robust enough for the most demanding enterprise tasks. You are expected to engage at the implementation level, contributing to architecture decisions alongside engineering rather than simply handing off requirements.
• Context Engineering:
  Own how our Agents manage the context window as a deliberately controlled resource. This includes progressive disclosure of tools and skills, context compaction and summarization, offloading of large payloads to a persistent file system, and the instrumentation that keeps agents oriented across extended trajectories.
• Model-Scaffolding Co-evolution:
  Own the feedback loop between North's harness and the Modeling Team. This PM is the connective tissue that makes that possible: ensuring harness design decisions are validated by Modeling before they are built, that evals are the shared bridge between both teams, and that as the harness evolves the model evolves with it.

• Define and own the roadmap for North's agent harness, including the agent loop, context engineering layer, tool orchestration, sandbox execution, and sub-agent delegation.
• Serve as the primary interface between North engineering and Cohere's Modeling team, ensuring new harness capabilities are validated before being built and that neither team paints itself into a corner.
• Own North's agentic evaluation framework, ensuring evals are compatible with both the North harness and Modeling's training infrastructure, and that they serve as a reliable bridge between product and research.
• Engage enterprise customers to surface real-world agentic failures and translate findings into concrete product and model requirements.
• Stay current with the open-source and commercial agent ecosystem and drive adoption decisions that keep North's architecture aligned with emerging standards.

• 5+ years of product management experience in agentic AI systems, developer infrastructure, or applied ML products.
• Deep understanding of modern LLM agent architectures, including multi-agent systems, tool-augmented reasoning, memory and retrieval, programmatic orchestration, RAG, and long-horizon execution.
• Strong grasp of agentic evaluation design, including how to measure task completion, failure recovery, and long-horizon reliability, and how to diagnose model vs. scaffolding gaps.
• Technically deep enough to contribute to architecture decisions at the implementation level: comfortable reviewing and shaping design docs, reasoning about async execution patterns, sandboxed environments, file system design, and the tradeoffs that come with building harness capabilities into a production platform.
• Ability to flex between ML research conversations and engineering architecture discussions with equal fluency.
• Track record of shipping platform-layer products with demonstrated impact on reliability, performance, or capability.

• An active practitioner of agent frameworks who regularly builds with and follows the latest developments in open-source harnesses, coding agents, and orchestration tools in both professional and personal work.
• Hands-on experience with enterprise agentic deployments: multi-tenant orchestration, tool permissioning,…