AI training pays software engineers more per hour than almost any other domain — typically $60–$150/hr for mid-level evaluation work, $150–$200+/hr for senior engineers at frontier labs, and the occasional FTE-equivalent ($300K+) for niche specializations. The work is part-time, fully remote, fits around a primary job, and the application process — while real — rewards the specifics most engineering resumes already contain.
This guide is for working software engineers (or new grads with portfolio work) deciding whether to take this on as a side income, and which platforms to apply to first.
What "AI training" means for engineers specifically
Five buckets, in rough order of pay floor → ceiling:
- Code RLHF. Compare two AI-generated solutions to a coding problem, pick the better one, explain why. The bread and butter of the category. $50–$100/hr typical, $120+ for harder languages (Rust, OCaml, embedded C) or systems-level work.
- Reference solution authoring. Given a coding prompt and a mediocre AI response, write the gold-standard solution the model should have produced. Harder and better-paid than comparison. $80–$150/hr typical.
- Code review of AI output for production scenarios. Read PR-shaped AI-generated code, find the bugs, evaluate the architecture, score on correctness, idiom, and maintainability. $100–$180/hr for senior engineers. Requires real production experience (the rubrics are written to be unscoreable without it).
- Specialty domain coding. ML engineering for ML model evaluation, infrastructure code for cloud platforms, security code review. $120–$200/hr. Same shape as the general tier, narrower talent pool, higher rates.
- Adversarial / red-team. Try to make the model produce insecure, broken, or harmful code. Document what worked. Often paid per-exploit ($50–$500 each) rather than hourly.
We've covered the underlying mechanic in our RLHF explainer — same feedback loop, just applied to code.
The five platforms hiring SWEs right now
1. Mercor — highest top-end pay
Mercor is where the highest hourly rates live. Senior software engineer review listings hit $130–$200/hr; FTE-equivalent senior engineering roles for frontier-lab clients top out around $300/hr at the ceiling. The application is the standard Mercor AI interview — see our Mercor review for prep. Pool: ~50 active listings at any time, ~20% of them engineering-focused.
2. micro1 — best for mid-level engineers and breadth
micro1 has the largest engineering listing pool — typically 60–80 active engineering roles at any given time across full-stack, mobile, ML, DevOps, and embedded. Pay is more compressed: $50–$120/hr typical, $150/hr at the top end. Acceptance bar is more realistic than Mercor's, and listings turn over faster, so it's the highest-volume first stop for working SWEs. Read our micro1 review for the application playbook.
3. xAI — tutor roles paying near FAANG-comp
xAI's Greenhouse board carries a steady stream of "tutor" listings for coding (Python tutor, JavaScript tutor, ML tutor) that are contract-based and pay $80–$150/hr. Easier path in than Mercor because xAI uses standard ATS application rather than an AI interview. Trade-off: project length is more variable.
4. Handshake AI Fellowship — best for new grads and PhDs
Handshake AI Fellowship is purpose-built for U.S. graduate students and PhDs. Up to $125/hr on part-time engineering and ML evaluation work. If you're a master's student or PhD in CS, EE, or applied math at a U.S. institution, this is structurally the most aligned program. ~200 active opportunities; the bar is real but credential-aligned.
5. Outlier — large pool, login-walled application
Outlier (Scale AI's contributor platform) is the highest-volume platform for code RLHF in the broader market — many tens of thousands of contributors. We don't ingest live listings (their platform is login-walled, see the "Pending Playwright" note in our best platforms guide), but it's worth applying directly at outlier.ai. Pay tends to be lower than Mercor / xAI (typically $25–$70/hr for coding tasks), but task pipeline is steady.
What makes an engineer pass the application
AI training platforms evaluate engineers on three signals, in this order of weight:
- Specific past work, talked about specifically. The AI interview will ask "Tell me about the most complex system you've shipped." A generic "I worked on a distributed system at Acme Corp" answer fails. "I built the Kafka-to-Postgres ingestion pipeline handling 30K events/sec, with retries through a Redis dead-letter queue, deployed across three regions" passes. The specificity is the signal — it's almost impossible to fake under follow-up probing.
- Range across the stack. Listings vary — frontend evaluation, backend code review, DB query optimization, ML training pipelines. Engineers who can pattern-match across the stack get matched to more listings. If you've only done one layer your whole career, you'll see fewer opportunities.
- Code quality on the qualification task. Most specialist listings add a scored coding exercise after the interview. Idiomatic, well-tested code with clear naming beats clever golfed solutions every time. The rubric scores maintainability and idiom heavily.
For the general application playbook (resume positioning, country eligibility, multi-platform parallelism), see our getting-accepted playbook.
Realistic income at 10–20 hours per week
Working software engineers typically run AI training as side income at 10–20 hrs/week alongside a primary job. At common rate bands:
- $70/hr × 15 hrs/wk: ~$4,500/month side income
- $100/hr × 15 hrs/wk: ~$6,500/month
- $150/hr × 10 hrs/wk: ~$6,500/month
- $200/hr × 10 hrs/wk: ~$8,700/month
Pay is hourly, weekly invoicing, paid in USD via Stripe or Wise. In most U.S. tax situations the work is 1099 contractor income, which means you handle self-employment tax — set aside 25–30% of gross and file quarterly estimated taxes.
Where to start
Apply to Mercor and micro1 in parallel. Run xAI applications through their Greenhouse board once you have one acceptance in hand (the second platform is much easier than the first because you have track record). On the listings page, filter by source to compare what's live right now and sort by pay descending.
For the broader landscape across all 12 platforms (not just SWE), see our 2026 best AI training platforms guide.