Bigger builds now start with a short round of questions and a plan you approve — then show you exactly what's happening while they run.
Asking for a slide deck, a report, a dashboard, or a deeper dig used to be a leap of faith. Dagny would propose a build straight away, and the bullets on that proposal card were lifted from internal notes — occasionally contradicting what you'd actually asked for. Plan mode changes the shape of that conversation. Before a first build gets proposed, Dagny now asks one to three focused scoping questions on a single card: audience, length, angle, whatever the request left open. Options are numbered and tappable, every question has a Something else… row if none of them fit, and answers go back together with one Send.
The plan you approve is now written for you. Each proposal carries two to five short bullets in your own language, describing what will actually be built — no more internal shorthand leaking onto the card. Revisions stay quick: once a proposal is on the table, "make it eight slides" goes straight through without another round of questions. And because scoping only helps where a card can render, conversations over Slack and SMS behave exactly as before. Nothing runs until you press Start.
Once a build is underway, the chat no longer goes quiet. A collapsible progress rail sits alongside your conversation and narrates the run as it happens — "Asking Dagny for material," "Rendering slides — cycle 2, 6 slides," "Design review — round 1," "Printing the PDF" — with a pulsing current step, a ticking elapsed clock, and a Cancel button within reach. When the result lands, the completed timeline stays put so you can see how it came together. The old static "Started" placeholder is gone.
We've also cleared up a few things behind the scenes. PDF report builds were previously invisible to the progress system — no banner, no updates, no sign of life until you reloaded the page — and they now report in like everything else. Chained and overlapping runs used to shadow one another, which meant a finished result could be withheld or a Cancel could point at the wrong job; progress tracking is now tied to the specific run you're watching, so each one reports its own status accurately. The status one-liner in the chat banner lights up for decks, reports, and dashboards too, not just research runs.
Together these two shifts close the loop on longer work: you shape it before it starts, and you can watch it happen instead of wondering.
Give plan mode a try on your next deck or report — we'd love to hear how the scoping questions land.