what is blink?

the slowest thing on my computer is me typing.

a lot of what i put on the screen is contextual continuation: the thread already tells me what to say back, and the agent already gave me the answer. by the end of the day i’ve typed thousands of words that any reasonable person in my seat would have typed the same way.

so blink is built on a simple bet: most replies aren’t original, and once a tool can see what you see, it can write them for you.

the idea isn’t new and the model isn’t the hard part. the hard part is making it fast and quiet enough that you forget it’s there.

right now this matters most for anyone working with agents. every agent is a thread that wants a response, and the response is usually a function of context the agent just gave you. running four claude code sessions at once stops being a context-switching tax and starts feeling closer to piloting.

it doesn’t stop there. the same loop fits most of what people do at a computer. email, messages, comments, reviews, posts. anywhere the reply is a function of what’s already on screen, and blink collapses it into a keystroke.

when this works, your day stops grinding to a halt. the agent gets unblocked, the slack thread moves on without you. your keyboard stops being the bottleneck on responses you’d already finished in your head, and your attention is free for the small number of things that actually require you.

the future i’m building toward is one where your computer operates with you. it watches what you’re doing, knows how you’d respond, and stays out of the way until you decide. the words you would have typed are already there, ready. typing becomes the mode you switch into when you actually have something new to say.

blink reads your screen and writes the rest. once you try it, you won’t go back.