MTS - Agent Research

About Etched

Etched is building hardware for frontier intelligence. We co-design chips, racks, software, and manufacturing to deliver best-in-class throughput and latency across both prefill and decode workloads. Our first products are heavily focused on inference. Backed by hundreds of millions from top-tier investors and staffed by leading engineers, Etched is redefining the infrastructure layer for the fastest growing industry in history.

Job Summary

As a research lead, you will set the research direction for this effort and spearhead the development of self-improving agents that reliably complete difficult, long-running engineering tasks. At Etched, this loop compounds: better agents help us build and optimize our systems faster, better systems expand our capacity to run more experiments, generating feedback and training data that further improve agent capabilities. Your job is to make this loop work in practice.

This requires strong research judgment. Where will better memory, context, or tooling unlock new capabilities? Which tasks benefit from structured workflows, and which require more open-ended autonomy? When do agent limitations call for changes to the harness, and when do they require custom models trained with supervised fine-tuning or reinforcement learning? You will set research priorities and design experiments that guide our investments in in-context learning, model training, and harness design.

You will work closely with engineers embedded across the company, with an initial focus on ASIC development, inference, and supercomputing. First, you will build the environments that let existing models do their best work through better tools, context, feedback, and in-context learning. Then, you will use the resulting trajectories and evaluations to develop specialized models where they can improve capability, reliability, or cost.

This is a hands-on research leadership role. You should be comfortable setting direction, writing and debugging code, and carrying an idea through to production. Research success means measurable improvements in the work Etched can accomplish, with disciplined use of compute.

Key Responsibilities

- Own the research agenda for agentic systems across Etched. Identify the most important bottlenecks, prioritize promising approaches, and decide when to scale.

- Work directly with domain engineers to understand real workflows, diagnose where agents fail, and turn those failures into research questions with concrete evaluations.

- Develop systems for long-running tasks, including memory, context management, tool use, planning, and coordination between agents. Determine which approaches generalize across teams and which require domain-specific methods.

- Build evaluations that measure correctness, reliability, and efficiency on real engineering tasks, and verify that gains transfer to production.

- Train custom models using SFT, RL, and distillation, with the agent harness in the training loop. Use execution trajectories and failure analysis to guide training data and reward design.

- Scale experiments to hundreds or thousands of concurrent agents while maintaining reproducibility, observability, and control over compute costs.

You may be a good fit if you have (Must-have qualifications)

- Strong research taste and a record of turning ambiguous problems into clear hypotheses, decisive experiments, and useful systems.

- Excellent engineering ability. You can move between research exploration, agent experimentation, low-level debugging, and production execution, and you take responsibility for systems working reliably.

- Experience building agents for complex, multi-step tasks, with a practical understanding of context, memory, tools, evaluation, and failure recovery.

- Hands-on experience with LLM post-training, including SFT and RL.

- A track record of making meaningful progress under compute and data constraints. You care about the quality of the experiment and the …