Software Engineer - AI Agent Evaluation (Remote)
- Role
- Engineering Management
- Experience
- Mid
- Employment
- Contract
The listing doesn't say where it hires from. Check the description or the employer's site before applying.
No BS summary
You'll create challenging tasks and evaluation criteria for AI coding agents within realistic simulated environments, including building virtual company environments, designing tasks, writing tests to verify agent solutions, and iterating based on QA feedback. This is project-based, not permanent employment, with pay up to $50/hr equivalent (~20 hours per task, your own schedule).
Required skills
Required languages
_Please submit your CV in English and indicate your level of English proficiency._
Mindrift connects specialists with project-based AI opportunities for leading tech companies, focused on testing, evaluating, and improving AI systems. **Participation is project-based, not permanent employment.**
**What this opportunity involves** We're building a dataset to evaluate AI coding agents - how well a model handles real-world developer tasks.
You'll create challenging tasks and evaluation criteria within realistic simulated environments:
- Build realistic developer environments - a virtual company with codebase, infrastructure, and context (tickets, docs, conversations) that forms a believable development history
- Design tasks from intermediate states of these environments - craft the prompt, define what "solved" means, and ensure the task is solvable by an AI agent
- Write tests that verify agent solutions - accept all valid approaches and reject incorrect ones, neither too strict nor too lenient
- Iterate on tasks and tests based on QA feedback - review agent solutions, analyze failures, and refine until the evaluation is fair and robust
**What this is NOT**
- Not data labeling
- Not prompt engineering
- Not writing code from scratch - the agent writes most of the code; you guide and evaluate
**What we look for**
- 5+ years in software development
- Core stack: Python (FastAPI), JavaScript/TypeScript (React), Docker, Postgres, Kafka, Redis
- Experience writing tests (functional, integration)
- English proficiency - B2+
**Why this is hard **
Frontier models are already good at coding. Creating a task that genuinely challenges the best models is non-trivial. You need to deeply understand where models fail and what scenarios reveal the difference between a good and a bad solution. Tasks have many valid solutions - writing tests that accept all correct solutions and reject incorrect ones is harder than it sounds.
**How it works **
Apply → Pass qualification(s) → Join a project → Complete tasks → Get paid
**Compensation **
Up to **$50/hr equivalent** , depending on level and pace. Tasks are estimated at ~20 hours each; you set your own schedule.
What you'll do
- Build realistic developer environments - a virtual company with codebase, infrastructure, and context (tickets, docs, conversations) that forms a believable development history
- Design tasks from intermediate states of these environments - craft the prompt, define what 'solved' means, and ensure the task is solvable by an AI agent
- Write tests that verify agent solutions - accept all valid approaches and reject incorrect ones, neither too strict nor too lenient
- Iterate on tasks and tests based on QA feedback - review agent solutions, analyze failures, and refine until the evaluation is fair and robust
What they require
- 5+ years in software development
- Experience writing tests (functional, integration)
- English proficiency - B2+
Benefits
- Remote
- Flexible schedule (you set your own schedule)
- Up to $50/hr equivalent compensation
- Project-based work
Toloka AI supports frontier model post-training by building domain-specific reinforcement learning environments, tasks, and evaluation frameworks designed by real practitioners. Mindrift, powered by Toloka, connects top domain experts with cutting-edge AI initiatives.