Now that we know how fast we can build from our last month's Figma Make-a-thon, let's talk about whether what we're building actually works for everyone.

We're all using AI design and code tools now; they're just part of how things get built. But here's the catch: what they spit out is rarely accessible by default. And honestly, it's not like these are brand new problems. It's the same accessibility stuff we've been dealing with for years; now it's coming from AI agents instead of people, and happening at a much bigger scale.

So in this session, Claire and Chris are going to unpack why AI output falls short on accessibility, what issues keep showing up again and again, and more importantly, what happens when you actually give these models the right context to work with.

They'll walk through two skills the Intopia team has built: one called intopia-web-accessibility, which pairs a knowledge base of acceptance criteria with code generation, and another for building accessible prototypes directly in Figma Make, the same tool from the Make-a-thon, now looked at through an accessibility lens.

You'll get to see both skills in action, and they won't sugarcoat it either; there's stuff context alone can't fix, and we'll talk about where humans still need to be in the loop.

They'll also share what they've learned from putting all this to the test, benchmark testing, guerrilla testing, real user testing, plus a peek at the evaluation suite they're currently building to measure things like accuracy, cost, speed, and token use across different models and skills.

If you came to the Make-a-thon, this is where that story continues. And if you didn't, no worries, you'll still walk away with something you can actually use.