Senior AI Engineer

San Francisco · Hybrid · Full-time $200,000–$250,000 USD annual base salary · 0.25%–0.5% equity About CommodityAI CommodityAI builds software and AI automation for physical commodity trading. We help teams turn information scattered across emails, documents and existing systems into reliable workflows for trade execution and operations. Our customers work with the energy, metals and agricultural products that move through the global economy. We recently raised our Series A and are growing our engineering team. You will work closely with the founders and customers, own meaningful parts of the product, and see the results of your work in real operations. The role We are looking for a senior engineer who can turn the capabilities of modern models into agents that do useful work with increasing autonomy. You will build the harnesses around those models: the tools, context, execution loops, state and feedback that let an agent carry a task through to completion. The goal is to help CommodityAI scale what our product and team can accomplish. You will prototype new capabilities quickly, put them in front of users, and use evals and production traces to decide what is ready to run autonomously and what needs more work. This is a hands-on engineering role with ownership from the first experiment through reliable production behavior. What you will do

• Build agent harnesses that give models the tools, context and execution environment to complete real tasks across our product and connected systems.

• Design how agents plan, take actions, maintain state, recover from failures and hand work back to a person when needed.

• Prototype new features with models and agents, test them against real customer problems, and turn promising experiments into shipped capabilities.

• Build evals around task completion and the quality of decisions and actions. Turn production failures into reproducible cases that improve the next iteration.

• Inspect traces and debug the whole system: prompts, context, tool interfaces, orchestration and model behavior.

• Increase useful autonomy over time, with clear permissions, checkpoints and human intervention where the workflow requires them.

• Experiment with new models and approaches, measuring whether they improve capability, reliability, latency and cost in our actual workflows.

• Create reusable tools and harness components so each new feature is easier to build and operate, and the team can support more customers without repeating manual work.

What we are looking for

• Strong software engineering skills and experience building working agent systems that take actions and complete tasks.

• A practical understanding of how harness design, context management and tool interfaces affect model performance.

• Experience creating evals, reading execution traces and systematically improving unreliable behavior.

• The ability to move quickly from an idea to a working feature, then make deliberate decisions about what to harden for production.

• Good judgment about autonomy: where an agent can act independently, how to bound its actions and when to involve a person.

• Product curiosity and ownership. You care whether the system solves the user's problem, and you stay with it after launch.

Useful experience

• Long-running agents, durable execution, state management, retries and recovery.

• Coding agents, browser or computer-use agents, or agents that work across multiple tools and APIs.

• Building evaluation harnesses, task environments or tools for inspecting and replaying agent runs.

• Shipping model-powered features on a small team and keeping pace with changes in model capabilities.

What success looks like You expand the set of tasks our agents can complete reliably, shorten the path from feature idea to working product, and build evidence that tells us when to grant more autonomy. The systems you build let CommodityAI serve more customers and handle more work with the same team. Interview process We start with either a 15-minute introductory conversation followed by a separate 1-hour technical screen, or a single 1-hour session with a 15-minute introduction and 45-minute technical exercise. Depending on what we still need to explore together, we may invite you to a second technical screen before an onsite. We will explain the next step and what to prepare before each interview.