Agentic AI Harness Architect - Moveworks
Job Description
What You Will Do
• Set the technical vision for the Moveworks agentic AI harness, including planning, tool selection, context management, memory, critique, reflection, adaptation, and recovery.
• Invent and prototype algorithms that help agents complete longer, more complex tasks with less supervision and stronger outcome reliability.
• Design model-agnostic abstractions for combining foundation models, specialized agents, enterprise tools, policies, and deterministic workflows.
• Develop approaches for agent memory and learning from execution traces, user feedback, and task outcomes while preserving enterprise security and privacy.
• Create rigorous benchmarks and evaluation methods for tool-use correctness, plan quality, task completion, groundedness, policy compliance, latency, and cost.
• Explore methods such as iterative planning, ReAct-style execution, self-critique, reflection, test-time computation, multi-agent coordination, and verifier-guided reasoning.
• Build high-fidelity prototypes, identify the ideas that merit investment, and work with platform and product teams to define a practical path to production.
• Partner across Agentic Systems, Search, Relevance, ML Infrastructure, Product, and Security to connect agent intelligence with enterprise knowledge, permissions, and actions.
• Remain hands-on with code and experiments while influencing architecture, mentoring senior engineers, and raising the technical bar across the organization.
Requirements
Department: Engineering, Infrastructure and Operations
Function: Other
Experience Level: Not Applicable