Founding Engineer: RL Environments, Evals & Agent Benchmarks | On-site San Francisco | Up to $300K + Equity + Profit Share

“We accept the reality of the world with which we’re presented.”

Christof, The Truman Show

Now imagine you built that world. Every email in the inbox, every record in the CRM, every file on the shared drive, every coworker who never actually existed. An AI agent moves in, works there for weeks, and never figures out that none of it is real.

That is the job. You build the town, you write the script, and you make sure that when the agent goes looking for the edge of the dome, it does not find a door painted on the sky. Then a frontier AI lab trains on what you built.

Our client, an early-stage AI agent company in San Francisco that is growing fast, is hiring two Founding Engineers to do exactly this. We have written this ad to be blunt about what they need, so you can decide in five minutes whether it is you.

The Role in One Breath

You design and build the environments, benchmarks and graders that frontier labs use to train and test AI agents.

Here is the twist that makes this role different: the raw material is real. Our client runs an agent platform that is live inside actual companies, handling ops, go-to-market and engineering work. That gives you a steady feed of production traces showing exactly where agents break. Most data teams are guessing where models fail. You will be looking at it.

Your job is to turn those real failures into environments and benchmarks that a lab will trust, pay for and train on. You report straight to the founders. You own each lab project from first scope to final delivery. Nobody hands you a ticket.

What You’ll Actually Build

• Whole synthetic worlds. A fully simulated company built from scratch: generated inboxes, records, documents and people, consistent enough for an agent to live and work inside for weeks.
• Fake software that holds its nerve. Mock inboxes, CRMs and internal tools that survive thousands of real agent tool calls without breaking character.
• Benchmarks from real failures. You dig through production agent traces, separate genuine failures from noise, and rebuild the real ones as tasks a lab will accept.
• Graders that cannot be gamed. Scoring that still works when the agent knows more than any human reviewer, hardened against reward hacking. When a model finds the loophole, you close it.
• Benchmarks that tell the truth. You check that a benchmark’s rankings line up with how agents actually perform in production, not just how well they perform on the test.
• Lab deliverables, start to finish. Handed a contract, you scope it, build it and ship what the lab needs. No hand-holding, no committee.
• The taste conversation. You work side by side with the founders on what separates a great environment from a useless one, and how labs really use this data in post-training.

The Non-Negotiables

This is the scorecard the hiring team will hold your application against. Be honest with yourself before you hit apply.

1. You have built evals, benchmarks, RL environments or training datasets for LLMs or agents. This is the dealbreaker. Industry, a research lab or an internship all count, but you must have built the thing, not just run someone else’s. Name what you built and tell us how you knew it measured the right thing.
2. 1 to 5 years of engineering or research experience. That means real work at an established organization, not a resume made up only of internships. Exceptional new grads with serious lab or research experience will get a look.
3. You have worked somewhere that lives and breathes this. An AI lab, an AI data or RL environment company, an AI agent startup, or a top research group. If your day job already involved agent traces, post-training data or graders, say so loudly.
4. A CS, math or related degree from a top-30 university. Not negotiable on this one.
5. You are AI-forward. You use AI coding tools every day, you have real opinions on agent products you have used, and you change how you work the moment something better shows up.

Bonus Points

None of these are required. Every one of them moves you up the pile.

• You found a real flaw in a widely used benchmark. Tell us which one and what was broken.
• You caught a model reward hacking your grader, and fixed it. Truman eventually sailed straight into the wall of the dome. Your agent found its own version of that wall. We want to hear how you patched it.
• You have been a founder, or gone through YC, South Park Commons, the Thiel Fellowship or something similar.
• You have a competitive spike. Math olympiads, competitive programming, debate, chess or other games at a high level, or high-level athletics. The client hires for a visible edge, so show yours.

Who Should Not Apply

Being direct saves everyone time. This role is not for you if:

• You think AI-assisted coding is cheating, or you would rather keep working the way you always have.
• You carry a grudge against particular AI labs. The labs are the customer.
• Your resume shows several short stints in a row with no clear reason behind them.
• Your background is mainly bio or life sciences rather than LLM and agent work.
• Your first question is about vacation days. It is a fair question. It just should not be the first one.

The Deal

• Base salary: up to $300K, depending on level. The band starts at $150K, and senior candidates land toward the top.
• Profit share: you take a large cut of the profit, after costs, on every lab project you deliver. Strong builders can earn well beyond base. Do great work, get paid for it.
• Equity: meaningful founding equity.
• Ownership: high autonomy, low politics. You make product and architecture calls yourself, without daily check-ins.
• Momentum: already generating serious revenue and growing double digits month over month.
• Location: on-site in San Francisco, Monday to Friday. Bay Area candidates preferred. No relocation support.
• Hours: arrive between 9 and 11am, leave when you are ready for dinner. Lunch and dinner are covered. There is no office on weekends, but expect to put in some work on one weekend day. The other day is all yours. Yes, this is a long-hours job. We promised to be direct.
• Visa: open to visa transfers such as H-1B transfers and OPT. Candidates who would need to move from overseas are not a fit.

How to Apply

Apply below with your resume and a short note, five lines is plenty, that answers one question:

Walk us through an eval, benchmark or RL environment you built. How did you know it measured the right thing?

That answer matters more than anything else you send. A sharp, specific answer beats a polished cover letter every time. Please also include your salary expectation, where you are based today, and your visa status.

Big Wave Digital is recruiting this role on behalf of our client. Every application is read by a real person, and strong profiles go in front of the founders fast.

In case we don’t see ya: good afternoon, good evening, and good night.