Overview

We're building AI agents for the domain where a single wrong digit has legal consequences. Every number has to be right. Every decision has to be traceable. If that sounds like the hardest version of the problem, it is — and that's exactly why we're building here.

What you'll build

Model-agnostic agent harness — structured reasoning loops, tool calling, safe code execution, parallel orchestration. Deterministic. Inspectable.
Distributed cloud runtime — hundreds to thousands of fully isolated concurrent agents. Container orchestration, ephemeral compute, filesystem virtualization, execution tracing.
Advanced harness systems — dynamic tool search, decision-time prompting, context pruning and compression.
Multi-layer memory architecture — working memory, episodic task memory, long-term semantic retrieval, compression over time

Who you are

4–8 years building distributed systems or AI/ML infrastructure
Strong instincts for correctness, determinism, and debuggability
You've shipped things with no playbook — 0-to-1, not feature work

Why this

One seat, not a headcount — you shape the architecture of everything from here
$3.3T global market, zero purpose-built AI infrastructure for it