"Drawing pictures of software is slightly idiotic"
A conversation with Brendan, co-founder of Dawn — an AI-native canvas trying to close the gap between intent and shipped UI.
A conversation with Brendan Kearns, co-founder of Dawn↗. Dawn↗ is an AI-native design canvas in beta, built by Brendan and his co-founder Leo in London.
You spent time at InVision before starting Dawn↗. That's a useful vantage point.
InVision had a real moment. For a while it was the dominant handoff and prototyping tool, the thing every design team was using. Then Figma↗ happened, and fairly quickly it didn't matter that InVision was well-built or well-funded. The competitive ground shifted and that was that. I watched a well-resourced team get disrupted before they could respond. You don't forget that. It makes you think carefully about whether the ground you're building on is stable.
What's the actual problem you're trying to solve with Dawn↗?
The problem with most AI design tools is that LLMs are sequence prediction machines. They're trained on text. They don't inherently understand space, hierarchy, or why one layout decision is better than another for a specific context. Most tools work around that with templates, which is fine, but it's not a real solution. We wanted to go a level lower and give AI genuine spatial reasoning about UI.
Our first attempt was a computational design engine, a middleware layer that could sit underneath any AI tool and give it that grounding. We shelved it. The compute cost was too high and the scope too broad. What we shipped instead is more focused: a canvas where you describe what you want, iterate in natural language, and get back structured, opinionated UI rather than generic output.
How do you know when it's working?
The test is whether it produces slop. That's the word we use internally. There's a huge amount of AI-generated UI in the world right now that looks like a design system threw up. We're trying to produce something with actual judgment in it, not just something that looks like a design. That's the harder problem.
Dawn↗ is deliberately not trying to own the whole stack. Why?
Because I watched vertical integration eat the tools I grew up using. Dawn↗ works alongside Figma↗, Cursor↗, whatever your team uses. It's agnostic. I think there's a version of this category where one tool tries to become everything and ends up doing nothing well. We'd rather do one thing well.
You said something I keep turning over: "Everyone has taste. Not everyone has the craft to execute it." Is that the whole thesis?
Pretty much. People know what they want to build. Figma↗ let designers ship what they could already see clearly. We're trying to do the same for everyone who builds products but never learned to design in the traditional sense.
What does that mean for designers?
I'll be direct about it: drawing pictures of software is slightly idiotic. That's a deliberate provocation, but I mean it. The specific job of producing high-fidelity Figma↗ files as a stand-in for intent, handed to a developer who then rebuilds it in code, that workflow is on borrowed time. Good judgment becomes more valuable as generation accelerates, not less. But the artifact changes. Anyone building a product becomes the designer of the end thing, not a specialist producing an intermediate representation of it.
Where does this go?
The third horizon, as I think about it, is software that designs and builds itself with humans in the loop. The current product is a long way from that. But it's useful to know what you're building toward. Right now we're focused on making the 99% of work that ends up on the cutting room floor faster and less painful to produce. That's where most of the friction lives.
Dawn↗ is at dawn.design. Brendan is based in London.