Staff AI Engineer, Agentic AI Harness & Quality - Moveworks

Description

The company wants an experienced engineer who has built AI-native products and platforms to help expand its customer-facing agentic AI features.

The team is eval-driven and data-centric. You will work with frontier LLMs, tools, and the infrastructure used to tune, evaluate, and serve AI agents in production, and you will get support from a large annotation group that builds accurate, inclusive, privacy-respecting datasets for tuning and evaluation.

Performance here covers output accuracy and quality, along with latency, reliability, and the behavior of the whole user-facing system. Engineers on the team enjoy designing and running high-performing compound AI systems.

The team moves quickly on hard product and engineering problems, aiming for more reliable agents and more value for customers. The work supports the goal of building the top enterprise assistant platform, and you will collaborate with other groups across the company.

Responsibilities

• Apply software and compound AI system engineering to deliver lasting value to all customers.
• Improve how well agents reach the goals users set, and reduce unwanted agent behavior in customer environments.
• Work on hard problems in AI harness engineering, including agent cognitive architecture, eval benchmark creation, context engineering, generative UI, dynamic orchestration, image and voice I/O, multilingual support, conversational memory, reasoning strategies, grounding and verifying generated text, deployment safety, and automatic harness optimization.
• Keep up with recent AI research, open-source code, and models, and build on them.
• Design scalable, adaptable solutions to difficult problems.
• Apply machine learning and LLM fundamentals to improve the harness and its AI components, test changes in small experiments, and bring working solutions to production at scale.
• Work closely with web and chat platform engineers to turn harness work into user experiences that impress customers.

Requirements

• You ship product improvements as production-quality code with full unit tests, and you rigorously evaluate changes to models, prompts, context curation code, and other tunable components.
• You understand LLM and agent evaluation well, including agent action trajectories, generated text, and non-uniform sampling.
• You care about detail and keep eval datasets high quality.
• You think carefully about API contracts and what exposing a platform feature means for different types of users.
• You can start working quickly on a Mac, writing Python and/or Golang with help from coding agents.
• You know common agent architectures and large language models.
• You read critically enough to spot subtle miscommunication bugs in English agent instructions, and you fix them with precise writing and evals.
• You move fast and take strong ownership of your work.
• You keep shipping production-grade improvements steadily and like finishing hard projects quickly.
• You stay curious about engineering beyond your own field and keep learning to stay current in agentic AI.

Conditions

Your work persona is flexible, remote, or required in office, assigned based on the nature of your work and your work location.

Eligibility for a work persona may be checked by measuring the distance between your primary residence and the nearest company office using a third-party service.