Stas Bilder's last name means "pictures" in German and "builder" in English. At Palo Alto Networks, he's a solutions architect for Prisma Browser, the company's cloud-delivered security platform used by enterprises worldwide. His official job is keeping it secure and tailoring it for customers. His other job is building the internal automation that keeps 3,000 colleagues moving.
He'd used Zapier before, at an earlier startup where he helped foreign companies expand into the US market — web forms connecting to CRMs, basic integration work. When he joined Palo Alto Networks, the habit came with him.
A problem that couldn't be scripted away
Prisma Browser started as a startup in Israel before being acquired by Palo Alto Networks. With a global company came a high volume of demo requests, and one channel everyone already used: Slack.
Every demo account, login code, and license reset came through that channel. Someone read it. Someone went and did the thing. "Traditionally, we were handling that all in a Slack channel," Stas says. "People were just sending us messages with requests, and we were processing these." At Palo Alto Networks' scale, with hundreds of employees demoing the product, the approach stopped scaling.
The problem wasn't just volume. It was variety. Employees asked for the same things in completely different ways — a formal request, a casual one, an extraordinarily polite one from a colleague in the Middle East: "Would the bot be so kind as to please reset my password and provide me with customization rights?"
A traditional script couldn't handle that kind of variation. "To process that with a traditional script would have taken forever," Stas says. AI Zap workflows handle that kind of variation naturally.
What he built and what it saved
Stas built the system in Zapier. Employees who need access to the demo environment type into the Slack channel they already use, and an AI Zap interprets the request regardless of how it's phrased. The Zap creates accounts, assigns licenses, grants customization rights across multiple systems, and sends a response back to Slack.
"Behind one request for a demo account, there's a lot of moving gears," he says. "Creating an account, assigning licenses, granting customization rights in several systems — they need to make customizations to the product to show it to their prospective clients."
He also built an approval flow that works the same way. When an employee tries to access a site that requires permission, the browser sends a request to Zapier, which formats it as an interactive Slack message. "From the Slack message, I can approve or decline interactively," Stas says. If the site is already approved, the AI handles it automatically. "I just got myself approved right from the Slack message without opening any internal systems. And it's all moving nicely and smoothly."
He A/B tested the responses going back to employees. Generic replies worked. Personalized ones worked better. "When the response allegedly comes from something more animated than a stupid machine, people pay more attention." 20% fewer follow-up questions. Same automation, different communication layer.
He went further. One spring evening, Stas decided the bot deserved a personality — somewhere between Marvin the Paranoid Android and Bender Bending Rodriguez. The first tests revealed something unexpected: a depressed, sarcastic super-intelligence wasn't a mode. It was the default. Politeness had to be enforced. After a week of chaining models so they'd correct each other, the bot reached a state best described as "safe to show to humans." Honest mode survived as an Easter egg — triggered only when the request includes a 🙂.
The system now serves more than 3,000 internal users. Stas calculated what it would cost to have a person do this work, trained on PAN's systems, processes, and edge cases. "We're literally saving one FTE for the company," he says. "So it's $150,000 a year, pretax."
Zapier as a prototyping layer
The access management system is the largest build, but not the only one. Stas also used Zapier to prototype internal process workflows — Proof of Concept automation, for instance. "Zapier is great for prototyping," he says. And he knows when the prototype is done.
By then the system had grown from about 3,000 internal users to five to seven thousand. It needed to run continuously and reach internal directories Zapier couldn't access. So he moved it. "We know exactly what we need to do," he says. "Let's hard-code it, and let's put it on the clock." It went to an Azure workload. The architecture was right; it just needed to run somewhere else.
What's next
The onboarding bot is about to stop being an internal tool. It worked. Employees responded warmly to the chat interface. Now Palo Alto Networks is considering building a version of it directly into Prisma Browser for customers.
"People like the human-like typing interface so much that we're going to productize it into the actual bot that comes with the browser," Stas says. "We're going to incorporate some of those parts in the browser itself. Building things is fun, but watching them work and help actual humans is the real reward. It's the kind of feedback loop most people in large corporations never close — they work on a bolt or a wire, but never see the whole machine run."
He's also published a how-to guide on the Zapier platform for colleagues at PAN who want to build similar workflows. One engineer's build, available for the rest to follow.
Stas Bilder is a solutions architect for Prisma Browser at Palo Alto Networks.
