boxxkiteFounding AI/Agent Infrastructure Engineer
EngineeringRemote (India)Full-time · Founding team
Build a real agentic project on top of boxxkite as your actual day job — a coding agent, an eval harness, an RL loop — and be the first person to hit every rough edge in the SDK, CLI, and tool surface before a customer does.
About boxxkite
boxxkite is a self-hostable, Kubernetes-native sandbox for AI agent code execution — one isolated pod per session, network access denied by default, with a framework-agnostic tool surface on top. The core is open source under Apache 2.0, and there's a hosted beta running the same code.
We run arbitrary, agent-generated code for a living, so the isolation model isn't a feature — it is the product, and it is our reputation. The security page and the repo's SECURITY.md are the honest version of where that stands today, including what we haven't solved.
Why this role exists
- Our customers are the people building agents: coding assistants, autonomous dev tools, eval harnesses.
- We don't actually know how well boxxkite works for them, because nobody here builds a real agent product on top of it day to day.
- We need someone who lives in that world to use boxxkite for real and tell us precisely where it falls short.
What you'll do
- Build a real agentic project on top of boxxkite — a coding agent, an eval harness, an RL training loop — as your day-to-day work, not a side demo.
- Be the first real customer for every new SDK feature, CLI command, and integration (LangChain, LangGraph, OpenAI-style function calling) before it ships.
- Feed back concrete friction: where the SDK API is awkward, where cold-start is too slow for your loop, where the tool surface is missing something a real agent needs.
- Help design the eval/benchmark and RL-environment surface — session recordings as eval seeds, reward-shaping hooks — from a practitioner's perspective, not a spec-writer's.
What we're looking for
- Real hands-on experience building agentic systems — coding agents, tool-calling loops, RAG pipelines, or RL environments.
- Comfortable in Python, familiar with at least one of LangChain/LangGraph or similar.
- Opinionated about what makes a sandbox good to build against — you've been burned by bad developer experience and can articulate exactly why.
- Bonus: a published or open-sourced agent project, a contribution to an agent framework, or evals/benchmarks you've built.
Compensation
Competitive cash, plus meaningful early-employee equity.
We benchmark cash against the top of the market for the level and location rather than against what we can get away with, and every founding hire gets a real equity grant. We'll put the specific band on the table in the first call, before you spend time on anything else.
How we hire
- 01
A 30-minute call
What you've built, what broke, what you'd do differently. No trivia, no whiteboard.
- 02
A take-home, 4–5 hours
Unpaid, and hard-capped at 4–5 hours precisely because it is. You'll do it against the real product — hosted signup or a self-hosted clone — not a toy repo. Five calendar days, and ask for more if you need it.
- 03
A walkthrough, 60 minutes
You drive. Two-thirds on what you built, one third on what you deliberately left out — the omissions are usually the interesting part.
- 04
A conversation with the founder
Comp, equity, what the first 90 days actually look like, and your questions.
How to apply
Send us something agentic you've actually built, ideally with a repo we can run.
Prefer email? Write to careers@boxxkite.com directly instead.
Curious what the take-home looks like? Go create a free account or clone the repo and run something in a sandbox. Every one of our assignments starts there.
Other open roles
Founding Infrastructure & Security Engineer
Own the isolation model end to end — and be the person who gets to say “that's not safe to ship”.
Security · Remote (India)
Founding Full-Stack Engineer
Own a whole product surface top to bottom, with no specialist at any layer to hand it off to.
Engineering · Remote (India)