Compound AI Agent Runtime Engineer (Founding-Team Interest, Equity Only)

Founding-Team Interest — Equity-Based Opportunity. i.Wonder Laboratories is currently identifying prospective founding-team members. Cash salaries have not yet been established. These listings are expressions of interest in future positions, not offers to begin unpaid employment. Any work arrangement will require a legally compliant agreement. Compound AI Agent Runtime Engineer (listed on our site as Agentic AI Engineer) at i.Wonder Laboratories: founding-team interest opportunity Build the runtime that lets Julius run long, multi-step missions on real codebases: plan them, dispatch them, interrupt them, recover them and prove them, with model providers that stay replaceable. About i.Wonder. i.Wonder Laboratories, Inc. is an early-stage company building two connected products. Julius is a compound AI agent that runs missions across three tiers: a Prompt Architect, a Coding Agent and an Independent Reviewer. All three tiers share Project Brain, a durable mission memory. Wonderland is a persistent multiplayer open world for AI creation, built in Unreal Engine, where people and their AI agents create, collaborate and build software together. Both are in active development. What we describe here is their intended design, not a report of finished features. What you'll own - Julius's compound-agent runtime: how a mission is scoped, dispatched to its tiers (Prompt Architect, Coding Agent, Independent Reviewer), executed, reviewed, repaired and closed. - Multi-agent orchestration, including budgets, permissions and termination. - Deterministic and probabilistic modules working together: state machines and durable execution around non-deterministic model calls. - Tool calling and external integrations, with effects that are exactly-once or safely repeatable. - Interruption and recovery: pause, cancel, redirect, resume, checkpoints, and resumption from Project Brain after a crash. - Runtime integration with Project Brain (mission state, evidence and history), tracing, repair loops and governance, including keeping the Independent Reviewer genuinely independent. - Runtime reliability: deterministic tests, fault injection, end-to-end mission tests tracked over time. What success looks like - Missions can be interrupted, resumed and recovered from crashes without manual repair. - Runtime behaviour is covered by deterministic and end-to-end mission tests. - Any failure can be diagnosed from Project Brain records and traces. You have - Strong production software engineering: testing, code review, operating services. - Hands-on experience building agentic systems or LLM applications that execute tools. - LLM orchestration: context management, structured outputs, routing, retries, and cost and latency trade-offs. - State machines, workflow engines or durable-execution concepts. - Testing non-deterministic systems. - Clear reasoning about failure modes and concurrency. Nice to have (none of these are required) - Integrating agent runtimes with memory systems. - Agentic coding tools or agent harnesses. - Durable-execution frameworks or event sourcing. - PostgreSQL. - Publications, open source or write-ups on agents, orchestration or agent testing. Compensation and terms Equity: Proposed equity: 0.60%, four-year vesting with a one-year cliff (proposed terms, subject to final written agreements). Cash: Cash salaries have not yet been established (see the disclosure above). No salary is offered or implied. Type / location / visa: Founding-Team Interest · Remote (United States) · visa sponsorship is not available. Future team members are expected to make a serious time commitment once employment arrangements are finalized. How to express interest. Apply here on Wellfound. You'll need a résumé, links to your work, and short answers to a few questions. A person reads every application. More about i.Wonder: iwonderlabs.com