A few weeks ago, I was contacted about an opportunity to do contract work related to AI. I hadn't gone looking for it, and my first reaction was pretty simple: Is this legitimate?
The rate was $70 an hour. The work was remote, and the idea was that I could work when I wanted. I figured if I could pick up a few hours here and there, it might be a nice supplement to the other work I was doing.
So I did some research and asked questions. There was a background check, contractor paperwork, confidentiality requirements and an onboarding process. Nothing about the offer itself immediately struck me as suspicious. Sure enough, it was legitimate. The company was real. The work was real. And when I completed eligible work, I really was paid the stated rate, so I decided to give it a shot.
The First Cracks
That's when I started to notice a few cracks, particularly in onboarding. Access to some of the key systems you needed to actually do the work was delayed, unclear or otherwise convoluted. One set of instructions might tell you that if you had completed certain steps and still didn't have access, you should wait several days. Then you could open the private Slack workspace and find different guidance or a different explanation for the same problem. There was a lot of confusion, and at times it felt like the infrastructure was still trying to catch up with the onboarding process itself.
Given the size of the project, I wasn't immediately alarmed. This was a massive operation with a huge number of contributors, so some growing pains seemed reasonable. For me, the whole thing was an experiment anyway. I wasn't relying on the income, and I figured I would give it some time and see how things played out.
I assumed things would settle down. But like any good experiment, eventually you have to look at what's actually happening instead of what you expected to happen.
Getting Qualified Became Its Own Job
Once I finally had access to the systems I needed to do the work, I started noticing a different set of problems. One of the biggest was that access to the work itself was conditional. Before you could get to certain types of paid tasks, you had to complete required training and assessments. That sounds reasonable enough in theory. The problem was how much of that qualification process there was, how often it changed and how convoluted some of it became.
New assessments would appear. Existing requirements would be replaced or expanded. Some took hours to complete. If you got something wrong, you could be required to review the answer, explain your original thinking, explain why the expected answer was different and acknowledge what you were supposed to take away from it before moving on. Doing that once or twice makes sense. Doing it repeatedly across a very large number of individual judgments becomes something else entirely, and before long, I was spending significantly more time on the onboarding and qualification work they required than I was actually doing the generalist work I had been brought in to do.
Even the guidance around that time could be contradictory. I would receive automated warnings that my onboarding hours were high, while project communications elsewhere told workers to disregard those warnings because the required training itself was pushing people over the usual limits. That kind of thing became increasingly common: one message would create a concern and another would tell you not to worry about the concern created by the first message.
Then There Was the Queue
And completing everything you were supposed to complete still didn't mean you could simply log in and work. The next hurdle was the queue.
You essentially had to put your name in line for access to tasks. Sometimes hundreds of people were waiting. If the queue was full, you couldn't even join it. If you did get in, you still had to wait your turn.
One Saturday, I finally managed to get into a queue. The estimated wait was roughly an hour and a half to three and a half hours, so I figured I could work around it.
I ended up waiting more than four hours.
The problem was that I wasn't really free during that time. My husband wanted to go out to the pool, get something to eat and do normal Saturday things, and I kept feeling like I needed to stay available because I was waiting for the work to come through. When it finally did, I got roughly 45 minutes of paid work.
That was one of the moments when the economics of the opportunity started to look very different to me. The $70 hourly rate was absolutely real. But I had just spent a large portion of my Saturday waiting, unpaid, for the opportunity to earn it for less than an hour.
Hurry Up to Slow Down
The best way I eventually found to describe the whole experience was "hurry up to slow down." There always seemed to be some new urgency. Complete another requirement. Meet another deadline. Log in because work is available. Get into the queue before it fills.
Then you would do exactly what was being asked and find yourself waiting again, discover that the queue was full, or find that the available work was already gone. On at least one occasion, I received a notice that tasks were available and logged in almost immediately. By the time I got there, the opportunity was essentially gone and hundreds of people were already waiting.
The Work Finally Arrived and the Pressure Flipped
Once you finally got the work, the pressure flipped in the opposite direction. You didn't necessarily know what kind of assignment you were going to receive. One task might be relatively straightforward. Another could be significantly more involved, with new instructions and a lot of material to evaluate. Meanwhile, the timer was already running. Sometimes you had less than an hour to absorb the instructions, understand the assignment, make a series of judgment calls and complete the work.
There was a process for releasing a task if you couldn't finish it, and you could still be paid for eligible time already spent. But based on the guidance and conversations I saw, releasing tasks was clearly something workers were expected to avoid making a habit of. There was a sense that it could count against you in some fashion, even if the exact impact wasn't always clear.
Not surprisingly, one of the recurring frustrations I saw among other workers was that there simply wasn't always enough time to complete the work comfortably. I felt that pressure myself. So you could go from waiting hours for the possibility of work to suddenly needing to be completely focused and productive the second a task appeared.
The Rules Kept Moving
The harder part was that the standards you were being asked to apply were not always clear or consistent. I can't disclose the specific company, project or proprietary training materials because of confidentiality obligations. But broadly speaking, the work required a lot of judgment about language, context, audience and subject matter.
The rationale behind those judgments could be difficult to follow. You might read an explanation for why one answer was considered correct, apply that same logic somewhere else and then encounter an example that seemed to be treated differently. At times it became difficult to tell whether you were supposed to rely on your own professional judgment, memorize the platform's preferred interpretation, or somehow do both.
That left you making fast decisions under standards that sometimes felt like a moving target. And because continued access to the work depended on quality, you weren't simply asking yourself what the best professional judgment was. You were also trying to determine what answer the system wanted from you. Over time, that became exhausting.
I had gone into the opportunity thinking I would be applying my experience and judgment. Instead, a surprising amount of effort went into trying to reverse-engineer rules that did not always appear to behave consistently.
At times, the qualification process also felt less like professional calibration and more like remediation. I had been brought in because of my experience, yet I could find myself repeatedly documenting why I had missed an individual answer, why the expected answer was different and what I would do differently going forward, even when the underlying standard was the part I found unclear. It could feel less like being a professional preparing for a project and more like being a student required to show your work on every mistake.
I Kept Thinking There Was a Code to Crack
Still, I kept going. Part of the reason was that I could see from the private project community that many of the issues weren't isolated to me. People were asking questions about access, training, grading, queues and changing requirements. That made me less likely to conclude that I simply didn't understand the work.
I also kept thinking that I was probably one requirement away from figuring the whole thing out. Maybe the next assessment would be the last one. Maybe once I passed it, the queues would open up. Maybe once I understood the standards well enough, I would finally have more predictable access to the work. I kept thinking there was a code to crack.
Eventually, I completed one of the larger assessments that had become an important requirement for continuing the work. I spent a lot of time on it and tried to be thorough, and I passed well.
I remember feeling genuinely relieved. I thought, finally, I understand what they want. Maybe now I can spend less time trying to decipher the process and more time actually doing the work I signed up to do. Instead, I spent the next several days still struggling to get into the queues.
Just When I Thought I Had Figured It Out
Then I received an email saying I was being released from the project.
There had been no warning that my performance was a problem. In fact, the email specifically said the decision was not a reflection of my performance and was instead related to changing research objectives. After all of the training, assessments, queues, waiting and effort to remain eligible, that was the end of it. My access to the work disappeared, and shortly afterward, I was removed from the private project Slack as well.
Apparently, I Wasn't Alone
What made the whole thing even stranger was realizing that I apparently wasn't alone. Afterward, I spent more time looking through public Reddit communities where people doing this kind of AI contract work share their experiences. I found people talking about many of the same things I had been dealing with: repeated assessments, shifting requirements, confusing grading standards, long waits for work, tight timers and abrupt offboarding.
Some described working for months or completing hundreds of tasks before being removed. Others talked about finishing required training only to face another new requirement. There were people frustrated by the subjective nature of the assessments and others who said they felt relieved when the whole thing was finally over because of how stressful the work had become.
Not everyone felt that way. Some people essentially took the position that if they were being paid for training, they were happy to keep taking the assessments. That's fair too.
And Reddit certainly isn't proof that every worker had the same experience. Anonymous feedback is still feedback, and it should be taken with a grain of salt. But I wasn't looking for proof that everyone had experienced exactly what I had. I was trying to figure out whether the messiness I had encountered was unique, and it clearly wasn't.
One comment I came across afterward stuck with me. Someone essentially asked why people would put up with this kind of treatment.
The interesting thing is that when you're in the middle of something like this, you don't necessarily think of it that way. It doesn't present itself all at once as one giant red flag. It's an onboarding problem you excuse because the project is huge. Then it's an instruction you assume will be clarified. Then it's another assessment you figure you might as well finish because you've already gotten this far. Then it's a queue you wait in because maybe this time you'll get a few hours of work. You keep assuming you're almost through the messy part. That's probably what kept me going as long as I did.
The Fine Print Isn't Always in the Contract
Looking back, I do wish I had spent more time researching what actual workers were saying before I started. I had done enough homework to satisfy myself that the opportunity was legitimate, and it was. What I hadn't really researched was the experience of working inside it, and that distinction matters.
The company can be real. The opportunity can be real. The pay can be real. None of those things necessarily tells you whether the experience itself is going to be worthwhile.
I'm also realistic enough to know that even if I had read some of those Reddit threads beforehand, I might still have tried it. You have to take other people's experiences with a grain of salt, and I probably would have wanted to see for myself. In that sense, I'm glad I treated the whole thing as an experiment. I made some money. I learned a lot. And I got a firsthand look at a type of AI contract work that I otherwise would never have experienced from the inside.
But I would approach another opportunity like this differently. I would spend less time focused on the headline hourly rate and more time looking at everything surrounding it: how reliably you can actually access paid work, how much unpaid waiting and retraining is built into getting there, how stable the rules turn out to be once you're inside them, and what the people who have already done the work are saying about it. There is always fine print somewhere. It just may not be in the contract. Sometimes you find it in the experiences people are sharing on Reddit.
The pay was real. The opportunity was real. What wasn't obvious from the outside was everything that came with it: the waiting, the training, the changing requirements, the unpaid time, the pressure and the constant uncertainty. That's what I'd pay attention to next time.
