Join the $1M AgentX-AgentBeats Competition: Build the Future of Agentic AI!
A friendly reminder about the AgentX-AgentBeats competition from Berkeley RDI, open to students and recent graduates who want hands-on agentic AI experience.
Most competition announcements land in an inbox and get skipped. This one is worth a second look, because it is free to enter, open to students and recent graduates, and built around the kind of hands-on work that actually teaches something, regardless of whether a team walks away with a prize.
What the competition actually is
The AgentX-AgentBeats Competition is hosted by Berkeley RDI, an AI group at UC Berkeley, and it builds on the Agentic AI MOOC community of more than 32,000 learners. It brings together builders, researchers, engineers, and AI enthusiasts from around the world with one goal: push the boundaries of agentic AI through real benchmarking and real agent-building work, not just a lecture series.
Two phases, one shared goal
The competition runs in two phases. Phase 1 asks participants to build or enhance benchmarks for agentic AI, the tests that measure how well an AI agent actually performs. Phase 2 asks participants to develop AI agents that excel on those benchmarks. Together, the two phases are meant to create high-quality, broad-coverage, realistic agent evaluations as shared public goods, building toward a community-driven ecosystem of agent evaluation benchmarks that are standardized, reproducible, and discoverable rather than scattered and one-off.
Why it is worth doing even without a prize
The case for joining is simple: doing is how you learn, more so than sitting through a class. The value of a project like this comes less from winning and more from the act of forming a team, building something real, and testing it against other people's work, the same kind of learning that comes from starting a company or shipping a real product rather than just completing coursework. With over $1 million in prizes, cloud credits, and API resources on the table, and more than 700 teams already signed up, there is also real upside if a team's work stands out. But the floor is what makes it worth doing: if a team puts in the time and does not win, they still walk away having learned a lot.
Who it is realistically for
This is aimed squarely at students, recent graduates, and AI professionals who want real-world experience with agentic AI systems. It does not require deep credentials to start, just a willingness to form a team and commit the time. Since it appears to be free to sign up, the practical barrier to entry is low, and the learning opportunity from either phase, benchmark-building or agent-building, is substantial on its own.
Key takeaways
- AgentX-AgentBeats is a two-phase competition hosted by Berkeley RDI, building on a 32,000-plus learner Agentic AI MOOC community.
- Phase 1 focuses on building or enhancing agentic AI benchmarks; Phase 2 focuses on building agents that perform well on them.
- Prizes, cloud credits, and API resources total over $1 million, with 700-plus teams already signed up.
- The competition appears to be free to enter, lowering the barrier for students and recent graduates.
- The real value is the hands-on experience of building and testing something real, whether or not a team wins a prize.
Try it yourself
If forming a team and committing the time sounds feasible, the competition details and sign-up are on the Berkeley RDI site. This is not a Humanitarians AI program, but the same principle drives a lot of what Humanitarians AI teaches its own Fellows: doing the work is how the learning actually happens.
Full transcript(auto-generated, with timestamps)
[0:01]Hey, Professor Barry here. I don't know how I got on these people's mailing list. Uh, but this looks like a nice thing for students and recent graduates. This is called the Agent X agent beats competition. There's something called Berkeley RDI. I don't know what that is. They're, you know, some AI group at UC Berkeley. Um, and it's like some massive 1 million in prizes agent thing. It's it's um you can watch the lecture. I don't think it cost anything to sign up. Um but it looks like a big thing. So it looks like this thing is done by you know sponsors. But take a look at this. So I'll I'll
[0:42]Put this. So this is for students. You know as you know I highly encourage you to do things more so than you even got to class. Doing is how you learn. So the more you can do while both you're a recent graduate and a student, the more you are going to learn. I'm also a student. I'm getting my MBA. I find that you know doing the classes is easy. The where you learn is when you start a company, you build something, you do something. So this looks like a good thing to do. So I highly encourage uh people to do it assuming it doesn't cost a ton. I haven't got into detail to
[1:20]Read this thing. I think it's free. So if it's free and you have the time to form a team to sort of do this, the worst thing is you learn a lot and you don't get prize. Looks like there a lot of little small prizes and some big big prizes. So I I encourage people to do this. So I'm just doing this to encourage people to do this. Okay, take care.
More videos
2:08Bridging the Pixel Gap in Browser Automation.
2:23How One Narrow Safety Rule Can Make an AI Less Safe Everywhere Else.
2:04Why splitting a chunk from its document makes it retrieve for the wrong question
4:20Three You Can Take Back. One You Can't.
2:21Why a 50-turn agent pays for the same screenshot 35 times unless it caches the pixels
1:53