Approval Is the New Click
Every generation of computing adds a new fundamental interaction, and everything else follows from it:
- The desktop era gave us the click.
- The mobile era gave us the swipe.
- The agent era is giving us the approval.
Not "approve this cookie banner" - I mean approve as in: the system did something you didn't explicitly ask for, and now you decide whether it lives or gets reverted. Tool calls that touch real APIs. Draft emails about to be sent. Refactors about to land. Purchases about to be made.
Signals I've been tracking in the morning AI digests:
- Anthropic's Claude can now terminate conversations with persistently abusive users - a system taking a unilateral UI action on its own behalf. Someone had to design what that looks like and feels like.
- The UK ICO flagged agents as the next compliance frontier - transparency and control requirements are coming for agent UIs whether designers like it or not.
- The Guardian ran a first-person piece on being interviewed by an AI assistant - and the word used was dehumanising. The interaction pattern is spreading faster than the interaction design.
Here's the uncomfortable part for anyone who's spent a career perfecting empty states and red(':')[1]: approval UI is not a modal. It needs provenance (why is the agent doing this?), reversibility (what happens when I say no?), and cost visibility (what did this tool call burn?). None of that maps cleanly to buttons and checkboxes we already have.
The teams who figure this out won't just ship better agent products - they'll define the vocabulary. Like it or not, "Accept all" is this decade's "I Agree."
Sources: morning AI briefings Oct 9–13, 2026 (Anthropic, UK ICO, Guardian, dev.to UI library survey).