Designing in the age of agents

Generating a screen used to be the hard part. Now an agent can produce a plausible layout in seconds — and so can the next agent, and the one after that. The hard part has quietly moved: it is keeping a team and a handful of agents pointed at the same evolving design, without anyone working from a stale copy.

A canvas, not a queue

Most AI design tools treat generation as a request-response queue: you prompt, you wait, you get a result in isolation. That works for one person and one idea. It breaks the moment two people — or two agents — need to react to each other’s work in real time.

M2 starts from the opposite assumption. Everything lives on one canvas:

• People and agents edit the same document at the same time.
• Changes stream in as they happen, with presence and cursors.
• The canvas is the source of truth — not a thread of prompts.

Why it matters

When the work is visible to everyone at once, review stops being a hand-off and becomes a conversation. An agent proposes a component; a designer nudges it; another agent wires it to the design system — all without leaving the surface.

That is the shift we are building for: not faster generation, but a real place to think together.