Product demo · 2 min 35
A real walkthrough of the running product — not a mockup. One plan, decomposed into work anyone can pick up. Then reality arrives: customer calls, your own observations, what your agents found in the codebase. Nothing changes the plan until you say so.
This is the plan for a nine-person AI startup that's betting the company on answer quality. Everything they're working on, in one place.
Three levels. At the top sit your priorities — the bets you're making. Each one breaks into work streams, and those into tasks anyone can pick up: design, research, operations, the policy writing nobody wants to own. Not just engineering. Some are for a person, some an agent takes straight off the board.
None of this came out of a planning workshop — it was decomposed from a few sentences about what the company is trying to do. And it doesn't have to live here either: if your plan is already in Jira or Linear, Nazr works with it where it is.
Then reality starts arriving. A customer call. Something an agent noticed while it was working in the codebase. An article someone flagged. It all lands here as signals — from GitHub, Slack, Notion, Granola.
No single one of them changes anything, and that's deliberate. They accumulate into themes, weighted by how strong the evidence actually is. Your own observation counts for more than an article somebody linked.
And when a pattern finally crosses the bar, Nazr proposes the change — as a diff you accept or dismiss. That's the only way the plan moves. Nothing rewrites your strategy behind your back.
You can feed it directly too. Command-K, anywhere — and you type it the way you'd say it. No project, no priority, no form.
A few seconds later it's proposed a task, picked the work stream, drafted why it matters — and logged the same thing as evidence. You just confirm.
Or you can just talk to it, and it works against the same plan.
And when the work starts, agents pull tasks over M-C-P and leave their working notes right on the task — so next week a different session picks up exactly where this one stopped. It goes the other way too: your codebase reports back what's actually built and what's still half-finished, so the plan is measured against reality instead of memory.