The pieces of an AI agent's screen, built to work together: answers that stream smoothly, tool calls you can open, approvals that can't happen by accident, and a plan that shows where the agent is.
4 components
The moment an AI agent asks before acting: what it wants to do, why, and exactly what will happen. Approve is held for a beat so it's never an accident; Deny is one press. Then it folds into a one-line record.
AI & Agents
An AI agent's plan as it works through it: the current step pulses, a step waiting on you turns amber, finished steps tick and the line fills down to the next, so you see where it is at a glance.
AI & Agents
An AI answer that arrives like writing: tokens come in bursts but words appear at a steady pace, each sharpening into place behind a caret, after a Thinking shimmer. Light markdown; sentences announced to screen readers.
AI & Agents
One step an AI agent took: the tool, what it sent and what came back. Queued, running with a light sweeping its edge, done with a self-drawing tick and the time taken, or failed. Opens to show arguments and the result.
AI & Agents
The agent wants to email your team on your behalf.
Plan