A customer sends "hi" on WhatsApp in the middle of the night. No app, no menu, no "press 1 to talk to an agent". At seven in the morning the business owner opens his phone and it's all there: the name, what the person needs, when for. Nobody designed a single screen for that to happen.
That's the idea behind Zero UI, the name Andy Goodman of the design consultancy Fjord gave in 2015 to a movement that was already underway: the best interface is the one you don't notice. He was talking about thermostats and sensors. I think the conversation got far more interesting now, with language models — and that's what I want to write about.
We stop asking "how do I design this screen?" and start asking "how do I solve this with no screen at all?". Sounds subtle. It isn't. It means design effort stops going into layout and starts going into three things: state (knowing what the user has already done and probably wants), decision rules (when the system acts on its own and when it asks), and trust (how the user audits what was done on their behalf).
This already exists in your life without you calling it anything. The car door that unlocks with the key in your pocket. The hallway light that comes on at 3 a.m. The bank transfer you authorize by speaking. The thermostat that learned your routine and stopped bothering you. All that time the industry called it "automation". The difference now: a language model understands the customer's loose sentence — "wanted a table for Friday, six people" — no form, no menu, no commands to memorize.
Three things. Language models got cheap and local: a 7-billion-parameter model runs on a consumer GPU and answers in two seconds. Sensors got everywhere: presence, position, voice — the environment knows a lot without being told. And culture turned: people are fine talking to assistants and letting systems act for them. The friction Zero UI demanded from users dropped.
Zero UI fails in silence. A button that doesn't respond screams. A system that decides wrong and says nothing screams louder — only later, after the damage is done. So the honest part of Zero UI is the log: the user must be able to see what the system decided, when, and why. Invisible cannot mean inaudible.
I test this in practice with the customer service setup I built for Data Service Market. A customer shows up, sends a loose sentence; the system understands the request, asks what's missing, returns price and scope, and hands the human only what matters. When it works, the customer never notices there was an interface. When it fails, they notice fast — and that's where the learning happens.
Zero UI isn't the future of screens. It's the future of problems: each one solved in the right place, which is sometimes a conversation, sometimes a sensor and, less and less often, a screen.
Term reference: Andy Goodman (Fjord), talk at Solid, 2015.