The control layer for AI agents.

I’m not writing this as a mission statement. I’m writing it because it’s true.

I have a child, and like a lot of people who think seriously about where AI is headed, I worry about the world she’s going to grow up in. Not in an abstract way — genuinely. AI is one of the best things to happen to humanity in my lifetime. It’s already changing how our students learn, how founders build, how people work. I believe that. I’ve built a company betting on it.

But a technology this powerful doesn’t get a pass on the ways it can go wrong. Models that lie confidently. Agents that hallucinate a plan and execute it anyway. Systems handed real permissions — shell access, browser control, API keys — with almost nobody watching what they actually do with that power. And underneath all of it, a harder question more serious people are asking every year: what happens as these systems get more capable than us, faster than we can build the tools to understand or correct them? I don’t think that’s science fiction. I think it’s a real possibility — deception, hallucination, and, at the far end, even extinction risk — and pretending otherwise is how we sleepwalk into it.

So we’re doing what little we can. explainx.ai started as, and remains, an education company — we’re not walking away from that, because teaching people to use AI well is part of the same fight. But we’re also expanding into AI safety and security, starting with Agentbeam: a product built to watch what AI agents actually do, and flag it when something looks dangerous, inconsistent, or dishonest. It’s not because we think one startup solves alignment. It’s because we know how to build software, we’ve spent years deep in this ecosystem, and this was the most obvious direction for that expertise to go.

This isn’t a pivot away from what we believe in. It’s the same belief, applied to the part of AI that scares us instead of only the part that excites us.

— Yash Thakker, Founder