Software Engineer, Agents

Union Square, New York Cityfull timeOn Site

Apply on Employer Site

mid

$175k–$300k

H1B: no

Mirage The Software Engineer, Agents designs and builds agentic systems to power creative workflows for short-form video. The role involves building end-to-end systems, designing agent architectures, integrating state-of-the-art models, and improving agent quality in production.

Responsibilities

This role focuses on designing and building agentic systems that power creative workflows for short-form video. You'll help shape how AI agents reduce friction, accelerate learning, and unlock new creative possibilities for our users — building agents that can take creative actions on a user's behalf — while experimenting with emerging models and techniques to bring these experiences to life in production.
Software Engineers on the agents team work across product, machine-learning, and systems engineering — owning meaningful problems end-to-end and helping define how our agents are built and deployed across the product.
While we don’t use hierarchical titles or levels internally, we are currently hiring engineers who have operated at senior or staff+ scope.
Build and ship end-to-end agentic systems that meaningfully improve creative workflows, balancing quality, performance, and reliability
Design agent architectures — how agents gather context, plan, select tools, and execute creative tasks reliably at scale
Integrate state-of-the-art models, combining internal research and external capabilities to power new agent experiences
Measure and improve agent quality in production, using experimentation, evaluation frameworks, and real-world feedback to guide iteration
What Makes You a Great Fit
5+ years of professional industry experience
A track record of designing and developing ML systems or agentic pipelines in production
Deep experience with context engineering: RAG systems, token optimization, context management at scale
Experience building evaluation systems and agentic infrastructure (context gathering, tool selection, multi-step planning)