Software Engineer, Agents

Mercor•New York, NY

3d•$130,000 - $500,000•Onsite

About The Position

Mercor is seeking a strong engineer to build scalable agentic products. This role involves working with a variety of technologies across backend, frontend, data, orchestration, agents/LLM, and observability. The engineer will be team-matched to either the Automation team, focused on building intelligent systems and agents to automate operational work, or the Studio team, which owns Mercor's evaluation system and annotation platform for RL environments and tasks. Mercor is a profitable Series C company valued at $10 billion, operating in-person five days a week in San Francisco, NYC, or London offices.

Requirements

Python
FastAPI
Django
Pydantic
Next.js
React
TypeScript
Tailwind
PostgreSQL
MySQL
Snowflake
DuckDB
Redis
Kubernetes
Temporal
Modal
Woz
LangGraph
LangChain
FastMCP
Harbor
NemoGym
Datadog
PostHog
LangSmith

Responsibilities

Own agentic features end-to-end — from scoping with researchers/ops partners through implementation, launch, and iteration on real customer feedback.
Design and ship LLM agents, harnesses, and verifiers — including the tools, prompts, and policies that make them reliable.
Build the Python/FastAPI services and Temporal/Modal pipelines that orchestrate agent runs, human-in-the-loop review and iterations.
Build state of the art RL environments that expand the capabilities of frontier agents, with realistic enterprise apps, simulated coworkers, and rich company data rooms that support tasks spanning hours to days.
Build tooling that turns agent trajectories into insight, from statistical analysis to automated failure mode detection.
Build and refine the full-stack surfaces and data infrastructure — craft Next.js/React interfaces where operators and experts work with agents, evolve data models to give agents the structured context and audit trails they need.
Define agent quality and drive continuous improvement — build evals, instrument traces, analyze failure modes, and iterate on prompts, tools, and guardrails while raising the bar for reliability, cost, latency, and UX.
Partner cross-functionally to shape agent autonomy — work with Product, Design, Research and Ops to draw the lines between autonomous action, propose-and-approve flows, and human-in-the-loop decisions.