5+ years in product, research, or engineering roles with a focus on LLM agents or AI systems in production.
Hands-on experience with frontier models and agents, including their behavior and evaluation.
Proficient in designing evaluations for open-ended model behavior and making product decisions based on them.
Daily user of AI coding tools like Claude Code or Codex, with fluency in reading code and prototyping.
Familiarity with ML experimentation and research workflows, able to engage in technical discussions with engineers.
Strong writing skills to translate ambiguous directions into actionable plans with measurable outcomes.
Demonstrated agency in problem-solving, taking ownership of challenges to deliver tangible results.
Responsibilities
Set the strategy and roadmap for agents and auto-research, prioritizing loops based on capability, cost, and speed.
Define criteria for load-bearing agents, including evaluations, reliability standards, and oversight mechanisms.
Transform promising auto-research loops into a production system with clear budgets and capabilities.
Own and measure key metrics like agent adoption and cost per result, ensuring they align with goals.
Prepare internal systems for agent readiness, enabling accountability across various functions.
Maintain model-agnostic practices to adapt to evolving workloads while preserving context and memory.
Prototype agents and workflows using AI coding tools, actively contributing to development rather than just planning.
Facilitate adoption of agents across teams, discerning when agents should not be trusted.
Benefits
Opportunity to work in a real-world test bed with direct impact on physical operations.
Engagement with agents as a core operational model, not just a side project.
Fast-paced environment where your contributions can rapidly influence company-wide practices.
Full Job Description
About the Role
Most agents are judged on benchmarks. Ours are judged on whether a 40-ton machine works better next week.
Bedrock is building a company that scales with data, compute and agents, so that each new capability, machine and decision costs less and lands faster than the last. Agents already do real work here, at scale: they triage field issues, assemble release evidence, draft plans and write most of our code. And we're already seeing how autonomous research can unlock new capabilities by leveraging the right data, evaluation and sim.
You'll own the product for this, end to end:
The agent platform. Where agents at Bedrock run, get evaluated, earn trust for load-bearing work, and get measured.
Robotics auto-research. Agent loops that improve our autonomy stack, our data, our simulation and our infrastructure, faster than a team of people could on its own.
You'll report to the Head of Product and work daily with our agent platform, ML, simulation and autonomy engineers.
What You'll Do
Set the strategy and roadmap for agents and auto-research at Bedrock. Decide which loops to build next based on how much they move capability, cost and speed.
Define what it takes for an agent to be load-bearing: the evals, reliability bar, human oversight and feedback loop a team needs before it relies on one.
Turn auto-research from a few promising loops into a production system, with budgets, verdicts, provenance and a clear line from experiment to a capability that ships.
Own the metrics that show whether this works, such as agent adoption, the share of capability work done by auto-research, and cost and time per result. Get them measured, then get them goaled.
Make our internal systems agent-ready so agents can read, act and be held accountable across data, simulation, code, planning and decisions.
Keep us model-agnostic as the frontier moves: the best model for each workload, while context, traces and memory stay ours.
Build. Prototype agents, evals and workflows yourself with AI coding tools instead of writing specs about them.
Help drive adoption across engineering, operations and commercial teams, and be the person who knows when an agent should not be trusted.
What We're Looking For
Required
5+ years in product, research or engineering roles, including shipping LLM agents or AI systems that people rely on in production.
Hands-on depth in how frontier models and agents behave: harnesses, tool use, context, evals, cost and failure modes. You form your own view of a new model within days of its release.
Experience designing evaluations for open-ended model behavior and making product calls from them.
AI-native in your own work. You use Claude Code, Codex or similar every day, read code fluently and ship prototypes.
Comfort with ML experimentation and research workflows: you can follow a research discussion with engineers and push on its rigor.
Clear, concise writing. You turn an ambiguous direction into a plan with measurable outcomes that a team can act on.
A track record of agency: you've taken a problem nobody owned and carried it to a real result.
Preferred
Experience at a frontier AI lab or AI-native startup working on agents, evals, RL environments or automated research.
Background in robotics, autonomy, simulation or ML infrastructure.
Built an internal developer or research platform adopted across many teams.
Published research or open-source work on agents, evals or AI for science.
Why This Role
A real-world test bed. Experiments close against physical ground truth from a live fleet, not a leaderboard.
Agents as the operating model, not a side project. Our tech scaling plan assumes agents and auto-research do a growing share of the work, and leadership measures it.
Scope and speed. You'll own a frontier problem at a company small enough that what you ship changes how everyone works within weeks.
