Careers
We do not hire by job title. If you are fascinated by AI agents, have your own views on recursive self-improvement, and can show work, we want to meet you.
Who we are looking for
Fascinated by AI agents
You have built agents yourself, hit the walls of tool calling, context, sandboxes and evals, and know the gap between a demo and something people use for years.
Opinions on RSI
You have read some of the recursive self-improvement literature, maybe let an agent edit itself. You care about how it avoids getting worse, not only how it changes.
Work to show
Code, papers, essays, products. Anything. We look at what you have made, not the keywords on a CV.
Clear writing
Remote work runs on text. Decisions, methods and retrospectives get written down. If it cannot be written clearly, it is probably not clear yet.
How we work
Remote-friendly
The company is in Shanghai, but you do not have to be. Remote teammates get the same trust and the same responsibility as people on site.
Boundaries first
Write down what happens on failure before building. Every step is reversible and recorded.
Built against real scenarios
Features come from customer sites and our own daily use, not from demos.
Small and direct
No layers. Problems are discussed directly. Your code and judgment go straight into the product.
What you might work on
Engineering, research and delivery all need people. What you do depends on what you are good at and what you want to do.
- The digital employee runtime: sandboxed execution, memory and notes, nightly consolidation.
- Night school evaluation and coaching: exams, judge calibration, regression gates and drift monitoring.
- Product UI: memory pages, nature and version history, line-by-line coaching review.
- On-prem delivery: deployment and acceptance on customer networks and DGX Spark clusters.
How to apply
Write to hr@createrole.com: who you are, what you have built, and your view on how an agent avoids getting worse as it changes. Include links to your work. We reply within two working days.