AI Engineer

Read this first

This is not a React job. It is not a front‑end job. It is not a job for someone whose résumé says "AI‑augmented" because they sometimes use Copilot.

We are hiring a builder whose entire practice has been rewired around coding agents. Someone who lives in Claude Code and Codex CLI the way other engineers live in VS Code. Someone who treats the terminal, the agent, and the harness as a single instrument. Someone who has felt — and can prove — what 10x development velocity looks like when the human is operating as architect, reviewer, and orchestrator rather than typist.

If that does not describe you, this is not the role. If it does, keep reading.

FLINT OS is a construction operating system built on Palantir Foundry. It is the connective tissue between site, asset, decision, and outcome for a vertically integrated design‑build general contractor.

This role lives at the edge of that system — the boundary where humans (superintendents, warranty teams, project leaders, mechanical VPs), agents, and the OS actually meet. The edge is where new value is created. It is also where the OS learns what it should become next.

Your job is to run the round trip between that edge and the architecture. Fast, repeatedly, with judgment. Every cycle should sharpen both ends.

Concretely, the round trip looks like this:

Over time, this work compounds. More edge experiments produce more behavioral and usage data. Better data exposes where friction lives and where primitives are missing. Sharper primitives let the next experiment start further forward. The OS becomes capable of proposing its own next interface — and eventually its own next workflow.

That is the meta‑layer. You are one of the people building it.

We are explicitly looking for a non‑traditional profile. The credentials we care about are not on Linked In.

You can describe a feature in natural language, decompose it correctly, hand the right slices to an agent, review the output critically, and converge on production code in a fraction of the time it would take to type it. You know when to let the agent run, when to interrupt, when to reset, and when to do it yourself. You have opinions about prompt structure, context window management, and tool use that you can defend.

You have meaningful production experience with at least one of Claude Code or OpenAI Codex CLI, and ideally both. You are comfortable in the terminal as your primary surface. You understand MCP, tool calling, agent loops, and the difference between an agent that helps and an agent that creates work. You can explain — concretely, with examples from your own work — how you have configured agents to operate inside a real codebase.

You think about agentic systems as systems. You can hold the shape of an end‑to‑end loop in your head — data, ontology, agent behavior, human decision point, feedback signal — and reason about where the leverage actually sits. UI is one expression of that thinking. So is a CLI tool, an evaluation harness, an agent capability, or a proposed change to the ontology.

You are comfortable picking the right surface for the problem rather than reflexively reaching for React.