AI Pulse · Under the Radar

Most AI tools still work like a hand-off. You give an agent a task, wait for an output, then decide what to change. doop is trying a different model, where the agent appears on the same shared design canvas as the human, with its own name and visible cursor.

It’s an open alternative to Paper.design that runs in a single Docker container. Agents connect through MCP, using OAuth, and can work directly on the canvas alongside people. The useful detail is the feedback loop. Rather than leaving a vague comment and hoping someone translates it into work, human feedback can flow back to the agent as structured tasks. That could make design collaboration more practical when a team is using several agents for different jobs, such as layout, copy, or visual variations.

There’s no need to overreact yet. The project is only one day old and is showing 122 stars a day, which is attention, not proof that businesses will adopt it. Still, the visible-cursor idea is worth watching. When people can see what an agent is doing, where it is working, and what it changed, review becomes far easier than reviewing a black-box output after the fact.

For operators, the bigger lesson is that AI work will increasingly need a shared place where tasks, progress, decisions, and human approval are visible. This is the kind of thing we build into an AI command centre, giving teams a clearer view of where agents are helping and where people still need to decide.