Alex Pezold spent two decades in cybersecurity before building and exiting a payment tokenization company for nine figures. So when he says he thinks about security first and hype second, it's not posturing; it’s coming from someone who's done the work.
Today he's CEO and co-founder of Agentech, a workflow automation platform built specifically for insurance claims. It's one of those industries that hasn't changed much in decades, runs almost entirely on manual processes, and has a labor problem that isn't going away. This makes it exactly the kind of place where AI can do real work, if you're willing to do it carefully.
Here are three things I took away from our conversation.
The Hardest Part of Selling AI Is the Board Room
Agentech is solving a real, expensive problem. Insurance carriers can't find enough people to process claims, and when a tornado touches down, they need to staff up in 48 hours. Automation isn't a nice-to-have, but a true necessity. Alex can walk into a room, show the ROI, address every privacy objection with his eyes closed, and still leave without a deal because someone's executive leadership simply won't sign off.
That's the part nobody talks about enough. Product-market fit is one thing, and organizational readiness is another. You can have the right answer and still lose to institutional inertia.
The flip side is just as interesting. Other carriers are all-in, specifically because the labor problem is so acute they can't wait. It’s the same technology, same pitch, completely different response — based almost entirely on how much pain they're already in. If there's a lesson there, it's that urgency is the real unlock. Features close deals. Pain gets people in the room.
Security Is the Foundation
Alex came up in cybersecurity, and it shows in how he's built Agentech. His take on most AI platforms? The security architecture is an afterthought, and that's a real problem, especially in a space like insurance where customer data is everywhere and regulators are paying attention.
The solution he's found that actually moves carriers past their hesitation: zero data retention endpoints. Data comes in, gets processed, and gets deleted. There’s no training on customer data and no lingering exposure. When he shows that (specifically the terms with the LLMs and how data flows through the system) the deals that look dead start moving again.
His broader point is one I think gets undersold in AI conversations: Securing AI isn't actually that different from securing any other technology. It deserves the same rigor, and most people aren't applying it. If you're building in this space and haven't thought hard about privacy by design, you're going to hit a wall with enterprise customers eventually. It’s better to build it in now.
Pick One New Tool a Week and Actually Learn It
When I asked Alex what advice he'd give to anyone trying to get more out of AI right now, he didn't say "pick the best platform." He said he adopts one new tool every week and actually commits to it. This is not the same as picking a quick prompt to see if it works. It’s paying for a month, using it for real tasks, and understanding where it breaks.
His current example: Gamma for building board decks and collateral. It's changed how he works. He never would have known that if he'd just stuck with what he already had.
His take on Claude Cowork, on the other hand? He tried it, wanted to like it, and kept spending more time managing broken connectors than actually getting work done. He called out early adoption without real reliability as a tax on your attention — and I've had the same experience. You can see the potential and still decide it's not worth the maintenance right now.
The broader principle is worth keeping: There's no one-size-fits-all AI stack. But if you're not actively experimenting, you're going to fall behind someone who is. The commitment to perpetual learning isn't just 'good' advice. At the pace this is moving, it's pretty much mandatory. As Alex put it: "It’s remarkable to live in this era and see what’s possible with technology."
Listen to the full episode of Actually Intelligent to hear more from Alex Pezold — including how Agentech went from being called Blink to getting a legal letter from Chubb, his thesis on why LLMs will eventually hit a data ceiling, and why he still thinks OpenClaude is worth watching despite the security concerns.
LEAVE A COMMENT