Agentic AI Engineer

Mission: Improve Julius mission execution, interruption handling, recovery, orchestration, state persistence, and runtime reliability.

  • Founding-Team Interest
  • Remote (United States)
  • 0.60% proposed equity
  • No visa sponsorship

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. Future team members are expected to make a serious time commitment once employment arrangements are finalized.

Apply online All roles

The role

Design and develop the agentic systems powering Julius, i.Wonder's Compound AI Agent. Build autonomous execution architectures, agent orchestration, tool-use systems, multi-agent coordination, deterministic and probabilistic modules, task planning, mission execution, recovery mechanisms and integrations with Wonderland.

This engineer focuses on HOW Julius operates, executes and coordinates intelligent agents.

Agentic AI Engineer and AI Engineer

Agentic AI Engineer builds the mechanisms that let Julius plan, coordinate agents, execute tools and complete missions. AI Engineer builds the underlying intelligence, retrieval, memory, model and evaluation systems that make Julius more capable.

See also: AI Engineer.

What you’ll do

  • Work on Julius’s runtime: how a mission is scoped, dispatched to its tiers, executed, reviewed, repaired and closed.
  • Handle interruptions: pausing, cancelling, redirecting and resuming missions without losing or corrupting state.
  • Build recovery: crash-safe execution, checkpoints, safely repeatable tool calls, and resumption from Project Brain after a failure.
  • Orchestrate the Prompt Architect, Coding Agent and Independent Reviewer tiers, including budgets, permissions and termination.
  • Persist mission state in Project Brain — evidence, decisions and history — so it stays consistent under concurrency.
  • Improve runtime reliability: tests, fault injection, observability and debugging of agent runs.

The technical challenges

  • Long-running, multi-step missions built on non-deterministic model calls and external tools.
  • Effects that are exactly-once or safely repeatable when tools act on real codebases.
  • Keeping the Independent Reviewer genuinely independent of the builder and answerable to evidence.
  • Provider-agnostic tiers: providers are replaceable, so the runtime must not depend on any single vendor.
  • Producing evidence: Julius’s architecture is research-grounded, and showing that it delivers better verified outcomes than simpler alternatives is part of the work.

What success looks like

  • Missions can be interrupted and resumed, and recover from crashes without manual repair.
  • Runtime behaviour is covered by deterministic tests and end-to-end mission tests whose results are tracked over time.
  • Failures can be diagnosed from Project Brain records and traces.

Required qualifications

  • Strong production software engineering, including testing, code review and operating services.
  • Hands-on experience building agentic AI 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, and persistence for long-running processes.
  • Experience designing tests for non-deterministic systems, such as end-to-end mission tests of agent runs.
  • Clear reasoning about failure modes and concurrency.

Preferred qualifications

None of these are required.

  • Integrating agent runtimes with memory systems: durable mission state, history and their trade-offs.
  • Experience with agentic coding tools or agent harnesses.
  • Durable-execution frameworks or event sourcing.
  • PostgreSQL.
  • Publications, open-source work or write-ups on agents, orchestration or agent testing.

About i.Wonder

i.Wonder Laboratories, Inc. builds multiplayer open-world AI creation platforms, and researches, develops and deploys Compound AI Systems and Compound AI Agents. Our thesis is that AI creation should not end when a chat window closes: people and Compound AI Agents should be able to build, collaborate, compete, create software and maintain what they make, together, over time, inside a persistent world.

Wonderland is our flagship platform: a persistent multiplayer open world for AI creation, designed so that AI-powered creation and real software development happen in the same place as collaboration, competition and play. Wonderland is in active development.

Julius is our flagship Compound AI Agent and the intelligence inside Wonderland. It governs three permanent tiers — a Prompt Architect (a large language model), a Coding Agent (an agentic coding system) and an Independent Reviewer (an independent agent harness) — and every tier reads from and writes to Project Brain, Julius’s durable memory. Work is finished when independent review and verification say so, not when an agent says it is.

Read the research papers · Watch the Wonderland simulation

What to know about this stage

i.Wonder is an early-stage company. Wonderland and Julius are in active development, and the descriptions of them here are their intended design, not a report of finished features. The research papers set out the architecture and the evidence still to be gathered.

How to apply

Apply online with the application form. You’ll upload your résumé (PDF or DOCX), add your LinkedIn profile and links to relevant work — for example shipped projects, repositories, a portfolio or writing — and answer two short questions about this role.