Mechanize, Inc.

We build reinforcement learning environments that frontier AI labs use to train and evaluate their coding models. In these environments, models carry out software engineering work such as building a feature, deploying an application, or debugging an issue in an unfamiliar codebase. An automated grader scores how well a model performed, and those scores become reward signals during training and measurements of what frontier models can and can’t yet do.

Frontier models are already surprisingly good at writing code. Our engineers find where they still break down and build environments that reveal those limits. Our current focus is software engineering, but our long-term goal is the full automation of valuable work across the economy.

Careers

We’re hiring software engineers to design and build these environments. Apply here.

Your job is to build software environments that frontier AI fails at. You’ll do most of the building through coding agents; writing code by hand is too slow now. The models are good enough that finding something they genuinely can’t do is not easy.

Learn how our interview process works, or read about what working here is like.

Essays

These essays explain our vision and how we think about AI, work, and the economy:

• Announcing Mechanize Inc.
• How to fully automate software engineering
• The upcoming GPT-3 moment for RL
• Sweatshop data is over
• Medical AI isn’t the bottleneck to medical progress
• Cheap RL tasks will waste compute
• The future of AI is already written
• Life after work
• Unfalsifiable stories of doom

---

Mechanize is backed by Nat Friedman and Daniel Gross, Patrick Collison, Dwarkesh Patel, Sholto Douglas, and Marcus Abramovitch.

For inquiries, you can reach us at our contact page. Press coverage is collected here.