AI Researcher, Agent Systems

Greylock Partners; San Francisco (CA); On-site; USD 150,000 - 210,000; Full time

Job summary

Greylock Partners is partnering with an early-stage AI company to explore AI agents in real company workflows and to build the systems needed to test their effectiveness.

You will work directly with the founders across research and engineering, focusing on agent environments, deployment, evaluation, and multimodal understanding. Model post-training is a possible future direction.

Qualifications

• Experience with agent evaluation, reinforcement-learning environments, simulation, or synthetic workflows.
• Experience with multimodal or video understanding to turn long recordings into useful representations.
• Research engineering that connects experiments to reliable, usable products.

Responsibilities

• Develop research approaches for understanding work from video, accessibility data, and other workflow context.
• Build or improve simulated environments that let agents practice and be evaluated before live deployment.
• Design evaluations that measure whether agents complete useful work reliably and know when to involve a person.
• Research how multiple agents can coordinate, operate safely and improve through deployment feedback.
• Build the software and experimental infrastructure needed to move promising ideas into working systems.
• Help define which research problems matter most as the company learns from customers.

Skills

Agent systems

Reinforcement learning

Multi-agent coordination

Research engineering

Job description

Greylock is partnering with an early-stage AI company focused on a gap between model capability and adoption. AI can already change how work gets done, but bringing agents into a real organization remains slow, manual, and difficult to scale beyond individual pilots.

The company is building a repeatable path from an initial AI-agent deployment to broader organizational impact. Its goal is to make those agents correct, trustworthy, robust, and scalable—and to help companies build on each successful deployment.

Summary

This is a research role for someone who wants to work on AI agents in real company workflows and build the systems needed to test whether they work.

Depending on your strengths, you could help create simulated company environments where agents can practice before deployment, develop ways to evaluate and improve multiple agents operating inside an organization, or turn raw video into representations that capture tasks, decisions, and handoffs.

You'll work directly with the founders across research and engineering, with close feedback from real deployments. The company is not training models today; its current research is centered on agent environments, deployment, evaluation, and multimodal understanding. Model post‑training is a possible future direction.

What You'll Own

• Develop research approaches for understanding work from video, accessibility data, and other workflow context
• Build or improve simulated environments that let agents practice and be evaluated before live deployment
• Design evaluations that measure whether agents complete useful work reliably and know when to involve a person
• Research how multiple agents can coordinate, operate safely and improve through deployment feedback
• Build the software and experimental infrastructure needed to move promising ideas into working systems
• Help define which research problems matter most as the company learns from customers

What We're Looking For

You may bring depth in one or more of these areas:

• Computer-use agents, tool use, planning, or multi-agent systems
• Agent evaluation, reinforcement-learning environments, simulation, or synthetic workflows
• Multimodal or video understanding, including ways to turn long recordings into useful representations
• Research engineering that connects experiments to reliable, usable products

You should be able to formulate an open-ended problem, build and test a solution, and learn from where it fails in practice. Research publications or frontier-lab experience are valuable, but neither is a substitute for strong technical judgment and hands‑on execution.

People who have built digital twins or sim‑to‑real systems in robotics or physical AI may also find the company‑simulation problem compelling. The key question is whether you want to apply those skills to software‑based work inside organizations.

About Us

Greylock is an early‑stage investor in companies including Airbnb, LinkedIn, Dropbox, Workday, Cloudera, Facebook, Instagram, Roblox, Coinbase, and Palo Alto Networks. Learn more at greylock.com.

How We Work

Greylock's Core Talent team provides free candidate referrals and introductions to our active investments. This posting is for direct employment with one of our portfolio companies. We review every application and reach out directly when we believe there's a strong potential fit.