Miro – Widget creator

Widget creator had to do two jobs at once. You can make a widget quickly for the board you're on, or build one you save so the whole team can use it across many boards. You describe an idea, try the result and keep adjusting it through chat. I explored those interactions in working prototypes, so we could test the whole experience and change direction quickly when something didn't work.

There was no earlier version to improve, so the widgets had to look good from the first one. I was part of creating the design prompt that tells the AI what a generated widget should look like. The aim was that a simple request like "Make a team poll widget" gives you a good starting point, without having to write a detailed brief.

I tested dozens of widgets built with Miro's widget SDK and used the problems I found to improve the instructions. To see whether a change actually helped, I built a side-by-side tool that runs different prompts on the same request, so we could compare the results instead of guessing.

The builder UI is built around helping people get started. The front shows example cards, a team poll, a sprint board, a who's next picker, a tug of war, so you can see what it can do instead of guessing from an empty prompt. I built these details myself in the production code and shipped them through reviewed pull requests.

At Miro Canvas 26, Jake Reed and Jeff Chow showed what teams can build with it. Jake made a candidate scorecard from a prompt, connected to company data and used together on the board. Jeff showed lighter ones, like the tug-of-war vote and the who's next picker.

I made those examples on purpose, including the playful ones. They fit back to icebreakers and play, and to where Miro's brand has been moving, a bit more playful and open. In the Canvas presentation we showed the more serious work versions first and ended on a high note with the playful ones, which shows the breadth of what you can make.

The real shift wasn't the interface. It was that Miro didn't have to invent every widget anymore. People could make their own, right on the board, for this week or for one odd thing only their team needs. The same idea fits board generation: a generated board can use the normal widgets like documents and shapes, or go further and make a custom one that fits what you asked for. That points to where I think the canvas fits in a world full of AI: as the place the work lands, stays visible and stays shared. I wrote more about that in A love letter to the infinite canvas.

I worked with Kseniia Gritsai, Jimmy Schementi, Yulia Mustaeva and many others.