Working at Dust

I joined Dust as a product designer a few weeks ago. My assumption going in was that I'd be designing AI interfaces. Chat windows, prompt inputs, the visual grammar of machine output.

What I'm actually designing is closer to plumbing.

That's not a complaint. Plumbing is underrated. Plumbing is what makes everything else possible.

The problem Dust is working on is specific and concrete and real: people spend a significant chunk of their workday switching between tools, re-explaining context they explained yesterday, and searching for information that definitely exists somewhere in the company but isn't findable when they need it. Dust connects into those tools and tries to surface the right thing at the right moment. The ambition isn't to replace how people work. It's to make the parts that are already working work better.

What surprises me is how hard that is to design well. When the product has to live inside Slack, Notion, a dozen different data sources simultaneously, you can't own the full experience. You're designing in the gaps. Sometimes the interface is invisible, and the only evidence it's working is that nothing feels broken.

Speed of feedback

The team ships something, watches how it lands, adjusts. I've been here a few weeks and already watched two features get completely rethought after real usage. Not through a long redesign process. More like: someone noticed a pattern in how users were struggling, the team talked about it, and it changed.

There's also an honest sense of what AI is actually good at right now, versus what it isn't. That instinct shows up in the design work too. Features get scoped until they're reliable rather than shipped at a size that makes them unpredictable.

The hard question

Being inside an AI company right now is strange. The discourse is relentless. Every week there's a new model, a new benchmark, someone declaring something dead. From the inside, the daily work looks nothing like the discourse. It looks like: why did this agent fail on this specific input. How do we write onboarding for someone who has never thought about context windows. What's the right error message when the AI genuinely doesn't know.

The hard design question isn't how to make AI impressive. Demos can be impressive. The real question is how to make it trustworthy enough that someone will actually delegate real work to it, and not feel the need to double-check every output.

I don't have a clean answer yet. But it's the most interesting question I've had to sit with in a while.