About HeyGen

At HeyGen, our mission is to make visual storytelling accessible to all. Over the last decade, visual content has become the preferred method of information creation, consumption, and retention. But the ability to create such content, in particular videos, continues to be costly and challenging to scale. Our ambition is to build technology that equips more people with the power to reach, captivate, and inspire audiences.
Learn more at www.heygen.com. Visit our Mission and Culture doc here.

About the Video Agent team

Video Agent turns a prompt, a document, or a handful of assets into a finished video. Behind a simple chat box sits a multi-step AI agent. It plans a storyboard, writes the script, chooses avatars, voices, and visuals, calls many tools along the way, and hands the result to our rendering pipeline. Customers use it every day to make training videos, product explainers, marketing content, and more.

We are a small team that owns the agent end to end. That covers the prompts and tools the agent uses, the backend services that run it, and the product screens where people chat with it and edit what it makes.

This is not a role where you only call someone else’s model through an API. HeyGen trains its own models in house, and the agent is built on top of them. You will work directly with those models and with the researchers who build them, which is an experience few applied AI roles offer.

About the role

You will make Video Agent better at doing what the user actually asked for, and you will prove it with data. Some weeks that means redesigning a tool the agent calls. Other weeks it means building an eval, running an A/B test, or shipping a new editing experience in React. You will own features from the first idea through launch and follow-up, and you will decide which problems belong to the model and which belong to plain, deterministic code.

What you’ll do

Build the agent. Design the tools, prompts, and context the agent works with. Improve how it plans, how it recovers from mistakes, and how it handles long, multi-turn conversations.
Measure quality. Build evals and offline replays of real sessions. Run experiments behind feature flags and read the results honestly. Watch real user sessions and agent traces to find where things go wrong.
Ship across the stack. Build user-facing features in TypeScript and React. Build backend services and long-running workflows in Python and Go.
Work closely with others. Partner with product managers, designers, and the research and rendering teams that the agent depends on.

What we’re looking for

You have built and shipped an AI agent that real people use. We mean a system where a model calls tools, plans across several steps, and acts on the results. A chat wrapper or a prompt demo is not enough on its own. We want to hear what broke in production and how you fixed it.
You know how to tell whether an AI system got better. You have built evals or test sets for model behavior. You can tell a real improvement from noise, and you are skeptical of results that look too good.
You are a strong full-stack engineer. You have 2 or more years of professional experience. You are fluent in TypeScript and React on the frontend and Python on the backend.
You have strong product sense. You use what you build, you care how it feels, and you look at how customers actually use it.
You take ownership. You do not wait for a spec. You find the most important problem, fix it, and tell people what changed.

A degree in computer science or a related field is welcome but not required. Equivalent experience counts.

Nice to have

You have fine-tuned models before.
You have worked on code-to-video models, which generate video by writing code such as HTML or animation scripts.
You have worked on video agent models in general.

Show us an agent you’ve built

With your application, please include a short note (a few sentences is fine) about an agent you have built. Tell us what it does, how you measured whether it worked, and one failure you found and fixed. A link to code, a demo, or a write-up is a bonus.