The Invisible Plumbing Transcript and Summary

When a visa mix-up forced Max Brodeur-Urbas back to his childhood bedroom, he started building a tool that could break the internet's invisible plumbing.

[Speaker 1]: In early 2024, Max Brodeur-Urbas had a problem. And not a small one. He was kicked out of the United States. [Speaker 2]: It was a visa mix-up, but the result was binary-he had to leave. So he goes back to Canada, moves into his childhood bedroom, and starts coding. He wants to build automation tools. But he realizes the tools that already exist-the giants like Zapier or Make-aren't built for the way he wants to work. [Speaker 1]: Right, because the old tools were built for logic. "If this happens, do that." But Max wanted to build workflows for AI-messy, unpredictable, creative work. So from that bedroom, he builds AgentHub, which eventually becomes Gumloop. [Speaker 2]: Fast forward to today, Gumloop has raised over twenty million dollars, including a massive Series A just last year. And it’s leading a new class of companies that are betting the old way of connecting the internet is obsolete. [Speaker 1]: Today on The Angle, we’re looking at the invisible plumbing of the internet. We’re asking whether the legacy pipes are about to burst, or if the old guard just pulled off the ultimate pivot. [Speaker 2]: And keep an eye on one specific detail as we go: a bill. We’ve seen reports of users running what they thought were simple AI workflows on the old platforms, only to wake up to monthly invoices climbing past four hundred dollars. [Speaker 1]: That number isn't just a billing error. It’s the signal that the entire economic model of automation is breaking. [Speaker 2]: It’s Sunday, February 22, 2026, and you’re listening to The Angle. [Speaker 1]: So to understand where we are right now, we have to look at how we spent the last decade. For about fifteen years, if you wanted two pieces of software to talk to each other-say, getting a lead from Facebook into a Google Sheet-you used a tool like Zapier. [Speaker 2]: It was the "glue" era. And it was strictly deterministic. You set up a trigger, and you set up an action. It was rigid. It was safe. It was "If This, Then That." [Speaker 1]: Exactly. But then the ground shifted. We moved from "moving data" to "reasoning with data." Suddenly, you didn't just want to move a file; you wanted a system to open the file, read it, decide if it was important, and then write an email about it. [Speaker 2]: And that shift created a split in the market. On one side, you still have those deterministic tools-Zapier, Make.com. They are fantastic at structured tasks. But on the other side, you have this new wave of probabilistic, AI-native tools like Gumloop or B3OS. [Speaker 1]: And the big question everyone is asking is whether these two worlds can coexist, or if the new guys are going to eat the incumbents' lunch. [Speaker 2]: Let's start with the argument for the new guys. We call this the Unbundling Thesis. The basic idea is that the old generalist tools physically break when you try to shove modern AI workloads through them. [Speaker 1]: Because the nature of the data changed. Zapier was built for structured data-rows and columns, JSON, predictable text fields. [Speaker 2]: Right. But AI work is unstructured. It’s scraping a website that changes every day. It’s reading a fifty-page PDF that’s formatted weirdly. If you try to build that workflow in a rigid, rules-based system, you end up with a mess of "If/Else" statements that’s impossible to maintain. [Speaker 1]: This is where Gumloop comes in. It treats the AI model…

Related coverage: Technology · Popular stories