Projects that kept us busy

This year, our community tackled a lot. Our AI Trainers completed 670,100 tasks across 218 projects! That’s hundreds of thousands of moments where real expertise met real AI challenges; where human knowledge helped shape smarter, safer, more capable systems. 

Some of these projects even pushed models into messy real-world territory: 

• Response Preferences Project: Professionals from complex fields like medicine, linguistics, and more designed real, multi-step dialogues rooted in their day-to-day work. The goal? Help models show their reasoning step-by-step. Scenarios came straight from classrooms, clinics, and offices, forcing the AI to think and communicate clearly while experts pinpointed strengths and weaknesses. 

• AI Safety Project: A red teaming project where trainers did something a little unusual: pretending to be bad actors to test model safety. It was tricky work, and it required creativity, subtlety, and a healthy sense of responsibility. By mimicking real-world strategies a malicious user might use, AI Trainers helped uncover blind spots and improve safety. 

Others stretched into creative territory. In our Creativity Project, trainers collaborated with a media client to improve AI-generated video quality. They judged motion, realism, and style, flagging what felt artificial and pushing toward higher standards. 

“I was amazed to see how AI handles video generation. This project helped me see the connection between text and visuals in a completely new way, inspired by both lifelike footage and surreal content,” said Sabina, a QA on the project. 

Although we completed over 200 projects, the insights our community gathered were immeasurable. Here’s what some of our AI Trainers had to say about their experiences this year:

“Diving into AI training, especially with reasoning projects, has made me even more curious about how AI fits into everyday life. Now, I can’t help but spot when AI makes reasoning mistakes, even outside of work! It’s been a genuinely rewarding journey.” Sunayana, Project Lead and QA

“Mindrift sharpened my skills in prompt engineering, data annotation, and quality assurance. I learned how to break down tasks into logical steps, identify edge cases, and ensure outputs meet high standards of accuracy and clarity.” Matthew, AI Trainer – Math

“It was like being a professor but the student was AI. I put together complex scenarios that would be hard for AI to solve, and then watched how it responded.” Fabienne, AI Trainer – Finance and Accounting

The community that made it all work

Mindrift wouldn’t be Mindrift without our incredible community. In 2025, we grew to 20,000+ AI Trainers across 20+ languages, with 70% holding advanced degrees. That’s a lot of curiosity, creativity, and expertise coming together.

This year, we tried to make it easier to connect, learn, and explore. We listened closely to our community of 685,000 followers on LinkedIn, running polls to see where more clarity was needed.

Community needs, straight from the source via LinkedIn.

You replied, we listened. In addition to creating even more helpful content, we came up with a few new resources like:

• Free AI Resources Library: Guides, free courses, riveting articles, and practical explanations anyone can use, all focused around AI and the tech industry. Check it out 

• “The Drift” Newsletter: Connecting and keeping our audience in the loop with highlights, updates, and community stories — every month, right in your inbox. Subscribe 

In addition to resources, the community itself made a huge difference. Trainers often praised our QA team for their guidance and support. 

“With this project, the first few tasks took days to complete and I found I was still not getting as close as I would have anticipated. The QA feedback was detailed and supportive, and with time and patience, there was a task where I finally knew what was needed. The first task that passed first time was cause for celebration, and after that, I was able to complete each within the suggested time requirement. It felt amazing!” said Naomi, AI Trainer. 

At the end of the day, our community is where learning, curiosity, and collaboration thrive, keeping us moving forward.

What changed behind the scenes

In 2025, we focused on making Mindrift smoother, clearer, and more fun to participate in. From feedback to new ways to contribute, here’s a look at some of the updates we pushed out this year. 

Quality reviews, done right

We gave quality reviews a refresh. Our biggest goal? Make task evaluations more transparent and consistent. Now, feedback is guided by shared quality criteria, giving AI Trainers better visibility into how their contributions are assessed over time. 

What this means for our community: Less confusion, clearer learning signals. 

Tasks you can count on

We introduced new task types with fixed rewards to boost flexibility and provide access to more opportunities for our AI Trainers. Straightforward, reliable, and easy to pick up alongside our time-based tasks — just the way “on your own schedule” should be.

What this means for our community: More reliability and opportunities to contribute. 

Self-service projects, made simple

We launched a new self-service project format, helping connect AI Trainers and clients directly. This means all project rules, payment details, and quality requirements will be found in the task guidelines provided by the client, while Mindrift will focus on quality control and support. 

What this means for our community: Faster access to tasks, clear guidelines, and a more streamlined experience. 

Bootcamp for confidence

We created an AI Bootcamp to make onboarding and skill-building easier than ever. Available right on the Mindrift platform, this self-paced, optional resource is designed to help AI Trainers better understand quality expectations and project workflows. 

What this means for our community: Build knowledge, get to know the platform, and stay confident as projects evolve.

Trends we couldn’t stop talking about

A year ago, we predicted the hottest trends for 2025. Now, we can confidently say this year was all about AI agents.

AI agents were everywhere — trainers, developers, and the AI community couldn’t stop talking about what they might mean for the future and how humans and AI collaborate.

Our LinkedIn community also weighed in and no surprise, AI agents were a big hit.

We couldn’t stop talking about them as well, so we created a short series of blog posts around the topic (with more to come next year!). First, we dug into what exactly an agent is, why they’re different from traditional models, and how they’re already showing up in creative workflows and everyday tools. It sparked a lot of conversation about where intelligence actually lives — and how trainers are uniquely positioned to shape its behavior.

We followed that up with the biggest AI agent stories making headlines. From governments and mega corporations experimenting with agents to in-depth research and expert opinions, we covered everything you need to know about AI agents right now. 

And because we love putting ideas into action, we introduced something brand new: agentic opportunities with Mindrift. These projects go beyond single tasks, focusing on helping to build and test autonomous AI agents. 

And we’re just getting started. If you want a sneak peek at what we’re watching for 2026, check out our latest trend roundup.