Software Engineer, Agent

About the Role

We're hiring engineers to build the agent layer behind OpenArt's next generation of creative products, starting with Director - which lets anyone "vibe direct" a multi-minute cinematic video by chatting with AI. You'll be building the agent infrastructure that pushes experiences like this further: longer-horizon planning, tool use across multiple model vendors, memory across sessions, and the harness that keeps all of it reliable in production.

This isn't a research role - it's a product-and-systems role. You'll build on and extend the agent harness (tool use, context management, multi-step planning, evaluation), develop the MCP servers and CLI tooling around it, and ship the frontend experiences that make all of it feel effortless rather than technical. You'll work closely with the founders, the agent team, and design to turn "what should this agent be able to do" into a system that actually does it - reliably, at scale.

What You'll Do

• Build and extend the agent harness that powers OpenArt's conversational and autonomous creative workflows: tool use, context/memory management, multi-step planning, and sub-agent orchestration.
• Build and maintain MCP servers and CLI-based agent tooling that connect our agents to internal services, external model vendors, and creative asset pipelines.
• Author reusable Agent Skills and structured workflows that let agents reliably handle complex, multi-step creative tasks - scene planning, storyboarding, asset generation, revision loops.
• Contribute to evaluation frameworks and observability so the team knows when an agent is working well, and can catch regressions before users do.
• Ship polished, production frontend experiences (React/Next.js) for agent-driven products, end to end.
• Make pragmatic trade-offs between speed, correctness, and scalability - and communicate them clearly.
• Partner closely with product, design, and GTM to translate user needs into technical solutions.
• Track the fast-moving agent ecosystem (MCP, Agent Skills, orchestration frameworks) and bring the best of it into OpenArt's stack.

Core Requirements

• 2+ years of full-stack engineering experience shipping production software.
• Hands-on experience building LLM-powered agents or applications that have reached real users - not just prototypes or demos.
• Working familiarity with agent harness concepts: tool use, context/memory management, multi-step planning, or sub-agent orchestration.
• Solid data modeling and schema design skills for real product use cases.
• A strong AI-assisted coding practice - you use tools like Claude Code, Codex, or Cursor as a core part of how you build, and have real opinions about what makes them work well.
• Product sense - you think about what makes an agent experience feel trustworthy and delightful, not just functional.
• Clear communication - you can explain agent behavior and technical trade-offs to teammates who aren't engineers.

Nice to Have

• Experience authoring reusable Agent Skills or structured agent workflows.
• Practical experience building MCP servers or CLI tooling for agents.
• Exposure to agent orchestration frameworks (Claude Agent SDK, LangGraph, or similar) or LLM evaluation / observability tooling.
• Hands-on experience with image / video generation models.
• Startup experience, or having built something 0 to 1 (side projects count).

Compensation

• $300k+ total compensation including base salary, bonus, equity.
• Equity - meaningful ownership in what you build.
• High autonomy, high growth environment.

Work Setup

• Bay Area preferred (hybrid available).
• Visa sponsorship available.

What Success Looks Like in 3-6 Months

• You've shipped a production agent workflow inside OpenArt's ecosystem, and real users are running it.
• You've meaningfully improved a piece of the harness, a set of Agent Skills, or the eval loop - and can point to the metric that moved.
• You own at least one part of the agent stack end to end and are the go-to person for it.
• You've formed a tight feedback loop with design, founders, and real users.
• You're proposing improvements and catching problems early, not just executing on specs.

If you want to build the agents that turn frontier models into creative tools millions of people use - and grow fast while doing it - we'd love to talk.