What Working an AI Training Contract Is Actually Like

Thang Tran

September 23, 2026

Slack, a corporate email, time tracking software, and pay every two weeks: the unglamorous day to day of an AI training contract, from someone doing them.

Onboarding: quick and soulless

Expect no humans. Onboarding at platforms like Mercor and micro1 is a chain of automated steps: identity verification, tax forms, tool signups. You will typically get a corporate work email for the project, a Slack workspace where all communication happens, and access to the work platform where tasks are assigned and submitted. There are usually hiccups, it is not a polished process, but having onboarded at both big and small tech companies as a software engineer, I would call it more streamlined than most corporate onboarding, precisely because no humans are involved.

The surveillance part, stated plainly

Many contracts require time tracking software, commonly Insightful (formerly Workpuls), installed on the machine you work from, tracking activity during billed hours. If you care about privacy, know this going in: it is standard for hourly contracts in this niche, and it is reasonable to use a separate machine or user account for contract work. Nobody checks where you physically work, but you must actually be a resident of wherever you claimed. That is what the identity checks enforce.

Getting paid

Payment runs through mainstream rails, Wise or Stripe, typically every two weeks, and tracking what you are owed is straightforward: hours logged in the tracker, tasks logged in the platform. Across my three contracts, payment arrived without drama, which matches the broad pattern of community reports.

The work itself

Think of it as the smarter cousin of data entry. You mostly examine something an AI produced and evaluate it: rate it, correct it, explain what is wrong with it. Or you produce a golden example for the model to learn from. The platforms usually allow AI assistance for parts of the workflow and absolutely forbid it for others; the rules are per task and they are serious about the forbidden parts. Some contracts are tedious, and then the next one is completely different. The variety between contracts is one of the genuinely good parts.

Who you are really working for

The platform is the middleman; the client is usually obvious from the project itself: the major frontier AI labs whose products you already use. Recognizing the client is also a legitimacy signal, because real lab work comes with real NDAs, real style guides, and real quality bars. Those NDAs are also why this guide names platforms, not clients.

Should you want this?

If you want a flexible, remote, decently paid contract you can run alongside a job or practice, and you can tolerate automated everything plus tracking software, this work delivers. If you need warmth, mentorship, or predictability, it will disappoint you. Check the live board to see what is open right now; every listing links to the official posting.