The Problem with AI

Just back from Web Summit ↗ (opens in a new tab) Vancouver. Ouf, what a week!

What stuck with me on the flight home wasn’t any specific company or pitch. It was the way the Web Summit Vancouver leadership curated the event with the selection of startups. Almost every conversation I had opened with the thing they were trying to fix, not the technology they were using to fix it.

This wasn’t the case when I went to Collision Conference in Toronto in 2023 - just a year after the ChatGPT hype burst onto the scene.

Just in my speaking slot, at Web Summit Vancouver, on the theme of “Building the Future Tech Ecosystem”, there were startups such as RoadToEV ↗ (opens in a new tab), working to make EV charging infrastructure more available across private and public charging systems. Weel Systems ↗ (opens in a new tab), helping pharmacists keep track of their logistics, systems of record and customer interactions. And, Mirantly ↗ (opens in a new tab), helping simplify the complex maze of legal immigration. AI showed up in basically all of those companies, just never first. It was the tool people picked up after they understood the problem deeply enough to know what kind of leverage they actually needed.

That was the exact motto at Amazon. Working backwards from a real human problem. You don’t start with “we have this new solution, who’d buy it?” You start with “here’s the customer’s pain, here’s the moment they hit it.” Then you reverse-engineer the thing that makes that moment go away. Most of what AWS got right in the years I was there came from that.

That’s the instinct we’re carrying into Sudden. Pain point obsession. Customer obsession one step earlier. CDN economics. The slow grind of a video platform’s CDN costs compounding year after year. The fact that video delivery architecture has been billing for the same unit of work forever.

We don’t get precious about WebRTC, or peer-assisted delivery, or any specific piece of the agentic infrastructure we ship today. Those are tools. WebRTC is the right tool for the problem right now. If something better comes along (and the internet’s protocols have rotated before, more than once), we’ll use it. The architecture is replaceable. The problem isn’t going anywhere.

Which is also why our name isn’t WebRTC-this or agentic AI-that. We’re not betting on a technology surviving. We’re betting on a problem getting worse. The CDN bill problem is older than peer-assisted delivery and it’ll exist long after.

Same logic is why the “AI for X” naming pattern makes me suspicious. Most “AI-powered something” companies are marrying a tool, not the problem.

And honestly, that’s the problem with AI itself. It’s a tool. A powerful one, but a tool. You still have to do the hard work first: find a problem worth solving, and find a group of people with enough incentive to actually want it solved for themselves.

AI hands you neither. It makes leverage cheaper once you’ve already done the thinking. It doesn’t do the thinking for you. And it sure doesn’t manufacture the demand. Skip either of those steps and what you have is a ticking time bomb.

Two questions I’d ask anyone evaluating us or any other vendor.

Founders: if you find yourself defending the technology in your company, ask whether you’re defending the problem or the tool. If it’s the tool, you have an exit problem.

Buyers: if you’re comparing two vendors that look similar on paper, ask which one is built around your pain and will evolve alongside you, and which one is selling you their tech.

Naming and pitch are a tell. The deeper signal is whether a team can explain the problem more clearly and obsessively than the technology they happen to be using today.

Pain in. Technology out. Tools change. The pain doesn’t.