AI Researcher (Agent Harness)

About Will

Wizard Intelligence Learning Lab (Will) is an entrepreneurial laboratory incubated by WizardQuant. Will is dedicated to building ASI for Sci-Tech. Bringing together exceptional AI talent, Will focuses on developing general-purpose AI systems for science and technology, with foundation models at the core of its work. Through an industrial-grade R&D approach, we advance model capabilities and explore their applications in depth, while pushing the frontier of AI across models, systems, and real-world use. What we seek to build is not a replacement for people, but a foundational capability platform that supports scientific discovery, technological innovation, and knowledge creation.

Job descriptions

The Agent Harness Team is dedicated to building agent execution systems capable of solving complex real-world tasks. We focus on core capabilities such as tool use, long-term memory, planning, error recovery, and multi-turn interaction, exploring how models and systems can work together effectively to complete open-ended tasks that may span hours, days, or even longer in realistic environments.

Our company’s vision is ASI for Science. The Agent Harness Team is responsible for building the core capability framework that supports this vision, while continuously validating and improving system performance through real scientific and engineering workflows. In parallel, we feed the data, failure cases, and operational insights gathered from real deployments back into model training, enabling continuous improvements in planning, tool use, and complex problem-solving. Ultimately, we aim to develop a new generation of research agents that can meaningfully assist, and eventually contribute to, scientific discovery itself.

Responsibilities

• Design core agent harness components, including planning loops, memory systems, tool interfaces, state management, and recovery mechanisms.
• Define the boundary between model intelligence and system orchestration to maximize reliability and efficiency.
• Improve agent performance on realistic long-horizon tasks such as multi-day software engineering, automated AI research, and frontier research.
• Translate deployment insights, failure modes, and behavioral observations into post-training improvements.
• Build robust infrastructure that supports scalable agent execution and experimentation.
• Design experiments, iterate quickly, and identify recurring behavioral patterns.

Qualifications

• Strong engineering and research experience in agent systems, LLM applications, or related areas.
• Experience building complex agents, orchestration frameworks, or execution systems.
• Deep understanding of model-system interaction boundaries.
• Familiarity with RL and RLHF concepts.
• Strong debugging, experimentation, and systems thinking skills.
• Experience with production agent platforms.
• Experience with memory architectures, tool ecosystems, and long-running workflows.
• Interest in both frontier research and foundational platform engineering.