SWE Coding Benchmarks & RL Environments

What you'll do

You build the problems that AI coding agents are trained and graded on. Each task is meant to look like something from a real team's backlog, set on a full-stack web app, and the codebase grows as each new task builds on the last. That means you are not writing isolated puzzles: you are extending a shared project and keeping it coherent.

The other half is the tests. You write the checks that decide whether an agent's solution is correct, so a vague or leaky test is a defect in the benchmark. The ad also wants you to go past the front end: new services and APIs, a backend in another language (Go, Rust or Java), gRPC and rate limiting, drawn from systems you have built. You are expected to say what has worked in other projects and flag what should change in this one.

The apps are TypeScript (React, Node, SQLite, Playwright). The ad explicitly allows using AI tools to write code and tests.

Who fits

• Solid professional software engineering experience on both backend and frontend
• You have built and debugged real systems and care about testing them well
• Clear written communication, and comfort working independently from written docs
• 10–20+ hours a week available, starting soon
• A LinkedIn profile and a GitHub account, which the ad says are required

US and Canada are preferred for timezone overlap, but this is a preference, not a residence requirement; the ad welcomes exceptional candidates elsewhere.

What it pays

$45–75/hour, paid weekly, on an ongoing contract. The ad publishes a range and does not say what decides where in it you land, so expect the work trial and your experience to matter. Terac's listing summary shows the top of the range, $75/hour.

Worth knowing

Good:

• A published hourly band with weekly pay
• Work that rewards real engineering judgment rather than rote labelling
• AI tools are allowed for writing code and tests
• Open worldwide, with only a timezone preference

Less good:

• The process has several stages: a short screener, a brief AI interview, then a paid work trial on the codebase you would keep working on
• 10–20+ hours a week is a real commitment for a contract role
• The stack is fixed (TypeScript web apps), so it suits you best if you are fluent in it
• The ad does not say how many people are being hired or how long the project runs
• Contract work: no benefits, and you manage your own tax

About this listing

Posted on Terac's public opportunities board on 19 September 2026 and read on 23 October 2026; the listing is set to expire around 18 November 2026. The pay, hours, stack and process are Terac's own text. The Apply button opens this study on Terac; you sign in or create a Terac account there to apply. Terac also runs lower-paid market research alongside this kind of work, so check the rate on each study. See Terac.