Software Engineer (RL Coding Environments)

What you'll do

You pick a repository you know well and build a task in it that a frontier AI model should struggle with: tracking down a subtle bug, building a non-trivial feature, refactoring a tangled module or making something measurably faster. Then you package it as a reinforcement learning environment.

Each task needs:

• A clear problem definition
• A reproducible environment the model can run in
• Deterministic verifiers that accept correct solutions across different valid implementations, while rejecting incorrect, incomplete or hard-coded ones
• A golden reference solution showing one way through

The appeal is the "your favourite repositories" framing: knowing a codebase well lets you find problems that are hard for real reasons.

Who fits

• Significant hands-on depth in at least one of Python 3, Java, Rust, Go, C++ or TypeScript
• Deep understanding of algorithms, data structures and software engineering principles
• Proven debugging of complex systems and effective optimisation work
• Experience refactoring large codebases or modernising legacy systems
• A record of shipping substantial features across teams
• Clear technical writing

Curiosity about AI is welcome; prior AI experience is not required. The card says "Software Engineer" while the ad body calls the role Senior Software Engineer, and the bar reads as senior.

What it pays

$50–100/hour is the band, but pay is output-based: you are paid per task that meets the specification, with a minimum weekly task count that the ad does not quantify. Time per task varies by experience and workflow, and building a trustworthy verifier for a performance or refactoring task is slow the first few times. Treat the band as a target and ask what a task typically takes.

Commitment

About 15 hours a week, contractor, remote. micro1 says it usually fills roles within 48 hours and expects first tasks within 24–48 hours of onboarding.

How it compares

This is close in spirit to micro1's Senior Software Engineer (Coding Tasks and Verifiers) listing, which also has you author tasks and verifiers at about 15 hours a week. The differences: this one starts at $50 rather than $30, is built around real repositories you choose, adds refactoring and performance problems, and does not single out security experience.

Worth knowing

Good:

• You work in codebases you already know
• A higher floor than the sibling task-authoring listing
• 100 openings
• Six languages accepted

Less good:

• Per-task pay with an unstated weekly minimum; rejected work is unpaid
• Verifier design is exacting and slow to learn
• Fast start expected after onboarding
• Contractor engagement: no benefits and your own tax to manage

About this listing

Posted by micro1 on its own jobs portal 18 days before we read it on 24 September 2026. The $50–100/hour band, 100 openings, 15-hour weekly guide and per-task compensation terms are its published text. No country restriction is stated and the customer is not named. See micro1.