Founding Engineer: Simulated Worlds, Reinforcement Learning & AI Agent Graders | San Francisco On-site | Up to $280K + Equity

"Welcome to the desert of the real." Morpheus, The Matrix

This is not a normal job ad. Read it like a puzzle.

SOMEWHERE IN SAN FRANCISCO

A company that does not exist is having a very busy week. Its inbox is overflowing. Its CRM is full of customers who were never born. Its VP of Finance has sent three passive-aggressive emails about Q3. None of them are real.

An AI agent has been working there for eleven days. It has no idea.

Someone built that company. Two someones, ideally. That is the job.

WHO WE ARE LOOKING FOR

You read an agent trace the way other people read a detective novel. When a model "succeeds" at a task, your first thought is: how did it cheat? You have, at least once, watched a model find a loophole in your grader and felt a strange mix of fury and respect. Then you closed the loophole.

You have built evals, benchmarks, reinforcement learning environments or training datasets for large language models or AI agents. Not used them. Built them. This one is not optional.

THE CLUES

The first two are required. The rest are how you skip the line.

• 1 to 5 years in engineering or research, somewhere that takes AI seriously: an AI lab, an AI data or reinforcement learning environment company, an agent startup, or a top research group
• A computer science, math or related degree from a top-30 university
• A spike: olympiad medals, competitive programming ranks, debate trophies, a chess rating, a sport played at a level that scared people
• You started something once, or went through YC, South Park Commons, the Thiel Fellowship or similar
• You found a crack in a famous benchmark and could not stop thinking about it

WHAT YOU WILL ACTUALLY DO

• Build entire fake companies: people, inboxes, files, office politics
• Build fake software convincing enough to survive real agent tool calls
• Hunt through real production agent failures (yes, real ones, from live customers) and turn them into benchmarks frontier labs will train on
• Write graders that still work when the agent is smarter than the person checking
• Make sure a leaderboard tells the truth
• Take a lab contract, scope it, build it, ship it. Nobody checks in on you every day

THE RULES OF THE GAME

• On-site in San Francisco, Monday to Friday
• Show up between 9 and 11, leave when dinner calls. Lunch and dinner are on the house
• One weekend day of light work. The other one is yours
• Bay Area preferred. No relocation support
• Visa transfers welcome (H-1B, OPT)

DO NOT ENTER IF

• You think AI-assisted coding is cheating
• You hold a grudge against an AI lab (they are the customers)
• Your resume reads like a series of short sprints
• Your first question is about vacation days

THE PRIZE

• Base salary up to $280K ($150K to $280K, depending on level)
• A big cut of the profit on every lab project you deliver. Good work pays twice
• Meaningful founding equity

HOW TO GET IN

Hit apply. Answer the questions. If you make it through, the first thing you will be asked is this: describe an eval, benchmark or reinforcement learning environment you built, and how you knew it measured the right thing.

Start thinking now.

Big Wave Digital is recruiting this role on behalf of our client. The client stays a mystery for now. Fitting, really.