The cost of switching context to use AI

You're in the middle of an email. You know more or less what you want to say, but the sentence isn't coming out right. So you stop, open a new tab, describe the problem to an AI, wait, copy the output back, and then edit it to sound like yourself.

That whole sequence took maybe 45 seconds. But something else happened too.

What a context switch actually does

Cognitive research on task-switching is pretty consistent: switching context doesn't just cost you the time it takes to switch. It costs you the time to re-engage with what you were doing before. Estimates vary, but 20-30 minutes of recovery for a deep interruption is a commonly cited range.

For a 45-second AI assist, that tax is usually smaller. But it's not zero. You left the document. You had to re-read to remember where you were. The sentence you were in the middle of? Gone. You're restarting.

Do this ten times in an afternoon and you haven't saved time. You've fractured it.

The promise that hasn't quite arrived

AI writing tools have gotten very good. The output quality is real. But the experience design has mostly assumed you're okay with stopping what you're doing to use them.

Chat interfaces are the dominant pattern. You describe what you want. The AI generates it. You copy. That's three deliberate steps, and each one pulls you away from the thing you were actually writing.

Side panels built into apps are a bit better. You don't have to leave the window. But you still have to describe what you need, wait, and copy. The mechanical friction is lower. The cognitive switch is still there.

When AI help doesn't interrupt you

The model that avoids this is autocomplete. Not AI you have to ask. AI that's already watching, already offering. You keep typing. If a suggestion fits, you take it. If it doesn't, you ignore it and keep going.

No tab switch. No prompt. No copy-paste. The workflow is still yours. The AI is a collaborator, not a detour.

That's the design principle behind Typeahead. A local AI that works anywhere you type on your Mac. No accounts. No cloud. No chat interface to navigate. Just suggestions, in place, as you write.

Why this matters more than you might think

There's a version of AI productivity that looks impressive in demos but doesn't actually change how work feels. You save a few minutes here and there. The baseline experience of writing is still interrupted and fragmented.

And then there's a version where the tool disappears into the work. Where you notice it working because the email came out faster, not because you had to stop writing to use it.

The second version is harder to build. But it's the only one that actually respects how attention works.