Software Engineering AI Training Jobs

AI labs pay software engineers to produce the material their coding models learn from. Software engineering AI training jobs are contract work: you write reference solutions, review model-generated code against a rubric, build environments an agent has to solve, and rank competing outputs. Almost all of it is remote, and most of it pays by the hour.

This page collects every live software engineering listing from Mercor, Outlier, DataAnnotation, Alignerr, micro1 and Terac into one feed, refreshed daily. Rates are shown as each marketplace publishes them.

155 open software engineering listings live right now.

FiltersActive
Clear all filters
Minimum pay
$25+ / hour$50+ / hour$75+ / hour$100+ / hour

Filtering by rate hides per-task listings, which have no hourly equivalent.

Category
Language & audio279Software engineering155Science & research145Finance & accounting128Business operations121Data & machine learning89Law87Medicine & healthcare83Cybersecurity67Engineering (other)59Arts & design51Sales & marketing35
Experience
Entry level48Mid level377Senior256Lead / principal29
Commitment
Contract / hourly1220Part time84Task based37Full time23
Arrangement
Remote1334Hybrid15On site3
Marketplace
Alignerr545Mercor412micro1360DataAnnotation19Outlier17Terac16
Date posted
Past weekPast monthPast 3 months

What the work looks like

The work splits into three tiers. At the base: grading model answers to coding prompts and fixing what the model got wrong. In the middle: writing hard problems with test suites, and reviewing pull requests the model produced. At the top: building reinforcement-learning environments, realistic codebases with tasks an agent has to complete, which is the closest this market gets to normal senior engineering work.

Seniority is the pricing axis. The marketplaces bid hardest for engineers who can judge production code, not just write it. A staff-level engineer reviewing agent trajectories earns a multiple of what a bootcamp graduate earns rating snippets, and the listings say so in the rate.

Where these listings come from

Six marketplaces publish this work openly. Every listing above links to the one that posted it, and each has its own page here covering pay, screening and who gets hired.

MercorOutlierDataAnnotationAlignerrmicro1Terac

Software engineering AI training jobs, answered

What do software engineers actually do in AI training work?

Four things, mostly. Write reference solutions to programming problems, with tests. Grade and correct model-generated code against a rubric. Review agent transcripts and mark where the reasoning went wrong. Build coding environments, small repos with a task and a verifier, that labs use to train agents. The mix varies by marketplace: DataAnnotation leans toward grading tasks, Mercor and micro1 toward project work.

How much do software engineering AI training jobs pay?

Published rates in this feed run from roughly $25 an hour for entry coding tasks to $150 and above for senior and staff-level project work. The middle of the market sits between $40 and $70. Rates track how hard you are to replace: anyone can rate a snippet, few people can review a distributed-systems design, and the listings price accordingly. Check the cards above for what is live right now.

Do I need machine learning experience?

No. The labs have ML researchers; what they are buying from you is engineering judgment their models do not have yet. Strong fundamentals, code review instincts and clear written English matter more than knowing how transformers work. The one exception is RL environment work, where familiarity with agent tooling and evaluation harnesses gets you into the best-paid projects.

Can I do this alongside a full-time engineering job?

Usually, yes. Most listings are part-time contracts with asynchronous hours, which is why the market skews toward working engineers moonlighting. Two things to check first: your employment agreement, since some employers restrict paid outside work, and the marketplace NDA, which will bar you from using anything you see in training data elsewhere.

How do these marketplaces screen engineers?

A coding assessment, an AI-run technical interview, or both. Mercor and micro1 interview you with an AI that probes your reasoning while you solve problems; Outlier and DataAnnotation gate access with unpaid assessments. The interview format trips up strong engineers who code silently: it scores how you explain decisions, not just whether the code runs. Practising that format before the real screen is the highest-leverage preparation.

The screen is where these roles are won

Every marketplace above screens with an interview or assessment in your own domain before assigning paid work. Run that interview with Skillora first and get feedback on how you explained your reasoning, not just on what you said.

Practise the screen free

Browse other fields

Data & machine learningMedicine & healthcareLawFinance & accountingLanguage & audioScience & researchBusiness operationsAll AI training jobs