Skip to content
NewEligibility scores are liveSee if you qualify

AI training jobs for software engineers: code evaluation explained

What coding AI projects ask engineers to do, from code review to building evaluation tasks, what they pay right now and how to pass the screening.

By Martin Flint, Talent & Community at Tier1 · Figures from , recalculated every hour

The short answer

AI labs pay software engineers to test and improve what their models say about their field. There are 112 such roles open on Tier1 right now. The median rate is $80 an hour, half of them pay between $65 and $110, and the best paid offers $263. The work is remote and usually part-time, designed to sit alongside a full-time job.

What coding AI projects ask you to do

  • Reviewing generated code. You read what a model wrote for a task and judge it: does it run, is it correct, is it how a good engineer would write it.
  • Writing reference solutions. Clean, tested code for a task, which the model learns from and is measured against.
  • Building evaluation tasks. Real-world tasks in a repository, with tests that decide whether a model's change actually fixes the issue. This is the kind of work behind coding benchmarks.
  • Ranking answers. Comparing two responses to the same request and explaining which is better, and why.

Who gets hired

Most roles name the languages and frameworks they need, and many ask for several years of professional experience. Seniority matters more here than in most fields: reviewing a model's code well means spotting what an experienced engineer would. Security engineers and data scientists have their own pages for cybersecurity roles and data science and ML roles.

How to stand out

  1. Lead with your stack. Name languages and frameworks with years, in the order you know them best.
  2. Point at code you wrote. A public repository or open source contributions do more than a list of employers.
  3. Show that you review well. Code review, mentoring and maintaining a codebase are the skills this work pays for.
  4. Expect a coding task. Screening is usually a short practical test. Write it like production code, with tests.

Working from outside the US

Of the 112 roles open now, 19 are open worldwide and 17 hire in the US only. The countries these roles name most often are United States, United Kingdom, Canada and Australia. The guide to working from outside the US covers the rest.

Current roles

Every open role is on its field page, with its rate, hours and the countries it hires in, updated every hour. To see how the pay compares with other fields, read how much AI trainers make.

Add your CV and Tier1 matches your stack, seniority, country and languages to every open coding role.

Join Tier1

More guides