About the role

This is not a prompting job. NinjaChat's agent already does real work over iMessage: it logs into websites and holds those sessions for weeks, watches prices and inboxes, runs scheduled jobs, executes code, remembers people across months, and acts through connected accounts like Gmail, Calendar, Notion, Slack, Linear and GitHub. Around 60 tools today, routed across 50+ models with mid-stream fallbacks so a provider outage never reaches a user. You own that loop end to end: new tools, the workflows that chain them into outcomes, the memory that makes it personal, and the evals that prove it works. We're building the ultimate AI agent, one that does everything for you. There's no research layer between you and production.

What you'll do

Ship new agent tools and the workflows that chain them into real outcomes
Own routing, fallbacks and latency across 50+ models
Build evals and harnesses that catch regressions before users do
Take new models from release post to production the same day

What we're looking for

You've built agents or tool-calling systems in production, not notebooks and demos
You think in workflows and failure modes, not prompts
You've handled provider outages, rate limits and silent model regressions in production
You benchmark before you believe anything, including your own work
Strong TypeScript, and you pick up anything else fast

Nice to have

You've built with MCP, Composio or similar connector ecosystems
Inference optimization or model serving

Apply for this role

AI Engineer · Remote