Build products the way you hand out free ice cream samples: Figma used a failed prototype to pick a direction, and an ugly demo to ignite company-wide collaboration.
Protoypes, failures, and unexpected breakthroughs ft. Noah Finer (Figma) | Config 2026 · Noah Finer
23 min total·Actually worth watching closely: ~17 min·3 must-watch clips
- 0:11 – 4:20Skim
Ice cream samples and the first version of Slides
Free ice cream samples as an analogy for the no-commitment value of prototypes, followed by the 2023 first version of Figma Slides: an infinite ocean of gray placeholders that almost no one could finish a deck with.
With several options in front of you, static comps — a pile of rectangles seen through a pane of glass — can't settle the decision; the first thing prototypes do is give the product direction.
The first half is a spoken analogy; the screenshot of the first version around 2:57 is worth a glance to feel how bad the gray ocean was. The rest you can just listen to.▶ Jump to 0:11Speaker · Noah Finer - 4:20 – 7:13Watch
Focusing on one slide: the hard problem on an infinite canvas
A before-and-after of the editor view once prototypes entered the picture, plus a demo of the 2023 flaw: on an infinite canvas, trying to stay on a single slide meant zooming out a bit too far dropped you back into grid view.
Spending a week or two on a low-fidelity prototype and putting it straight in front of real users is a cheap way to validate a direction — even a prototype that fails completely gives you the confidence to make the right call.
The 4:21 before-and-after and the 6:10 zoom-out demo are both about dynamic interaction; a description alone won't tell you where the flaw is.▶ Jump to 4:20Speaker · Noah Finer - 7:13 – 10:20Watch
Two failed prototypes produce a third idea
Both 'focus on one slide' candidates tested badly, but going through them produced a third idea: divide the viewport spacing by the canvas scale so it becomes dynamic. People liked it, and it shipped.
Prototypes aren't only for choosing between candidates — making them turns up better options than the ones on the table; skipping validation, by contrast, produces real user frustration.
From 7:44 he demos the dynamic spacing in action; why it's better is only explainable once you see the interaction.▶ Jump to 7:13Speaker · Noah Finer - 10:20 – 14:17Watch
FigPals: an ugly prototype ignites a snowball
Three weeks to launch with about one person's capacity, an ugly prototype proved pals could sync across accounts and set off self-organized collaboration: one person added a button, another added four pals, another brought them into the tool belt — around 30 people in the end, and 1.6 million FigPals created by users in two weeks.
A rough demo that actually runs turns bystanders into participants. Features were filtered by feel before launch: the distracting walking animation got cut, while the pal facing the direction it moves stayed.
The densest demo stretch in the talk — the ugly prototype in its original state, features stacking up, the animation trade-offs — is purely visual. Listening isn't enough.▶ Jump to 10:20Speaker · Noah Finer - 14:17 – 17:40Watch
LLMs speed up prototyping: four reordering options tested
For keyboard reordering — a 1D representation and a 2D grid that have to be kept constantly in sync — the models helped build four interactive prototypes fast for the team to try by hand; the winner, which never changes the rows of the unselected slides, shipped.
LLMs make prototyping far cheaper, but why the good option is good only becomes understandable after you feel it — deciding between prototypes is still on people.
From 15:43, the interaction differences between the four prototypes are the core evidence for the whole argument; 30 seconds of demo shows what diagrams can't.▶ Jump to 14:17Speaker · Noah Finer - 17:40 – 21:20Watch
A prototype relay across levels, and the limits of LLMs
An internal playground plus a Slack thread set off a relay: the PM, a designer, and even the CPO each built a placement for speaker notes, and a question open for over two years got solved and shipped through dozens of prototypes. Meanwhile, a model's read on a product experience stays ice cream behind a pane of glass.
LLMs let non-engineers prototype, but they can't feel it — their output is worth putting into the discussion, not treating as the answer. And validating a direction isn't the same as being releasable.
From 17:46 the PM's and the CPO's prototypes have actual screens worth seeing; the later argument about what a model can't taste is fine to just listen to.▶ Jump to 17:40Speaker · Noah Finer - 21:20 – 22:40Listen
Closing: prototype anything, embrace the mess
Prototyping goes all the way down to zero code — before writing any, fake the product with a slide deck, have people click through, iterate four or five times, then build. A bunch of rectangles and masks can pre-check whether a complex animation is even doable. Design is messy; make more prototypes and keep it fun.
Rather than chasing a tidy process, make 'build something people can try' the default habit — the bar for a prototype can be as low as one slide.
The close is mostly spoken summary with no key visuals; fine to finish on a commute.▶ Jump to 21:20Speaker · Noah Finer