Yuhki Yamashita (CPO Figma): Inside How Figma Built FigJam AI
Plus how PMs can work with designers and how to balance new and power user needs
Dear subscribers,
Today, I want to share an inside look at how Figma built FigJam AI, its new AI feature to help teams generate, sort, and summarize ideas.
Yuhki Yamashita is the CPO of Figma. I sat down with him to discuss how Figma:
Created a culture where PMs and designers love their craft
Built FigJam AI to solve real problems
Plans to make design accessible to everyone
How to create a culture where PMs and designers love their craft
Welcome Yuhki! To kick things off, how do you think PMs and designers can be great partners for each other?
For PMs, I would say:
Think about the constraints that led to the design. Instead of just critiquing a design, think about whether a designer is operating under the wrong constraints. For example, maybe they have to stick to a design system or what is feasible.
Speak the user’s language. PMs are often focused on metrics while designers are more human-centered. So talk to a designer about what the user might be feeling or what they might be trying to accomplish in their user journey.
For designers, talk proactively about the goals and trade-offs. PMs are often thinking about these so you might as well bring them up yourself.
I highly recommend reading more of Julie Zhou’s writing on this topic.
One of Figma’s values is “love your craft.” How do you look for this value in PMs during the hiring process?
I think PMs who value craft have the following traits:
Attention to detail. They would highlight small details that users care about in past products or in our homework assignment.
Excitement for great experiences. They inherently want to make the user experience better, even if it might not grow a metric.
Emphasis on the user’s POV. They naturally put themselves in the user’s shoes and feel the user’s emotions.
How do Figma PMs balance craft with moving metrics or optics like having a great product review?
Hopefully, it’s not just about having a great product review!
When it comes to metrics, I think every product team needs goals, but not all of these goals are measurable.
So when I evaluate PMs, it’s really about whether they’re setting good goals and achieving them. Goals don’t have to be metrics. For example, you can set a goal to solve a particular user problem or address a particular sentiment from the community.
So are there cases where you rely on qualitative feedback to understand if a problem has been addressed vs. running an A/B test?
Definitely. There are often downsides to running experiments. For example, an experiment might create new types of primitives that we wouldn’t want to support forever. Instead of experiments, we might:
Run betas to get feedback before making a launch decision.
Invest in products that we have conviction about.
Spending a lot of time with users goes a long way toward building conviction.
Overall, we try to be responsible and use the right measurement tool for the product.
You’ve also written about how PMs need to be responsible for “why” a particular problem needs to be solved. Do you have any tips for folks to become better at storytelling and explaining the “why”?
Put yourself in the shoes of the listener of your story.
I often have to force myself to start again from a blank slate to avoid making assumptions that the listener might not know about.
For example, listeners at a company all hands have significantly less context about your product than your immediate team.
You have to adjust your story for your audience.
Just to make this even more tactical, how do you prepare to explain the “why” when you’re making a big speech at a conference like Figma Config?
I spend a lot of time on the storyboard to make sure that everything flows together. It’s not just words but also visuals that create cadence and rhythm.
To be honest, I spend more time crafting the story than practicing its delivery. But that's just my style – I want to make sure that the frames in a storyboard are right and in the correct order.
You’ve been at Figma for over 4 years. How do you keep the focus on craft alive as the company continues to scale?
One of the questions that we ask every candidate and Figma employee is:
How are you a maker?
We like to hire people who have an itch to create things or tinker with side projects.
Communication and alignment indeed become more important as a company scales. But ultimately, I think people work at Figma because they love to make things and get their hands dirty. I’m that way and our CEO, Dylan, is too.
Do you have any particular rituals to celebrate this maker culture?
We have Maker Week twice a year where the only requirement is that you make something. People have made everything from prototypes to hand-painted murals. It’s an important week to pull folks out of their day jobs and give them a chance to think outside the box.
I think it’s so important for everyone to exercise the creative part of their brain.
Over the years, several features from Maker Week have made it to our roadmap, including FigJam AI.
How Figma built FigJam AI to solve real problems
I’d love to do a deep dive on FigJam AI next. Can you start by sharing some background on how FigJam evolved from Figma in the first place?
Back in 2020, it was the height of the pandemic. We started seeing people use Figma to connect through online happy hours, games, and more. They loved seeing the cursors of their friends and teammates interacting in real time.
That inspired us to think about how we could build a separate product for everyone in a team to collaborate (two-thirds of our weekly actives are non-designers).
We launched FigJam in 2021 as a digital whiteboard for teams. Since then, it’s been cool to see people use it for brainstorms, meetings, and all kinds of lightweight interactions.
What I love about FigJam is all the fun little touches like stamps and emotes. Was “fun” a core product principle early on?
Yes, I distinctly remember this moment:
We had a meeting with our board two months before we were going to launch FigJam. The board asked us what our differentiator was since there were many other whiteboard tools.
And Dylan said, “Well, it’s fun!”
But then we realized that the product wasn’t fun enough. So the FigJam team ran a sprint called “FunJam” to come up with all the playful features that you see today like cursor chat, emotes, and more.
So you’re absolutely right – fun was a core principle. Most workplace tools are a little boring and confine you in a box. We want to give you an inviting canvas with many lightweight ways to express yourself.
When it comes to metrics, it turns out that fun also increases engagement as well!
Amazing! So let’s talk about FigJam AI. How did this project get started?
We had a big AI hackathon earlier this year where a team built Jambot, an AI-powered widget that can transform a FigJam sticky into a full user story or even a haiku. We loved how it let people visually explore their ideas.
That motivated us to think more about the problems that people face in FigJam, such as how intimidating it can be to start from a blank canvas.
Emily and Jenny (the product and design leads for FigJam) are great at making FigJam boards. How do we deploy their expertise to every FigJam customer? We realized that with AI, we can actually do this. We know what a great FigJam meeting looks like – whether it is a weekly sync or project brainstorm.
AI can help generate the right template on the fly.
Yep, I’ve seen designers run amazing FigJam sessions which ironically makes me feel more intimidated to run my own. How did you come up with the feature to summarize and sort a sea of stickies in one click?
A brainstorming session can be engaging and collaborative, but it can also be messy.
We often end up with 100+ stickies that have to be manually organized and clustered. So we decided to use AI to:
Sort stickies into key themes
Summarize stickies into key takeaways and next steps
I want to highlight that all of these user problems existed in FigJam independent of AI. AI just happens to be an effective way to solve them.
As you know, building an AI product is often a long winding path. Can you share a particularly challenging moment in creating FigJam AI?
There were many such moments!
For example, we would look at early outputs of a FigJam AI summary from stickies – except the bullet points weren’t very helpful. Or we would look at an AI-generated template and realize that it wasn’t much better than just manually creating one.
With AI, PMs and everyone else on the team need to really own quality.
It’s easy to look at AI as a black box and say: “Well I don’t have too much control over this output.” But if you own quality, then you have to hold the bar high and say, “This is the kind of output we want and what we have right now is not good enough.”
You have to be willing to modify the prompt and what you’re feeding the AI to make it work. A lot of the PMs who worked on this project went through this process, including myself.
Yeah, from my experience building an AI product, it simply requires a lot of manual work. You have to be willing to modify the prompt and inputs, see the output, and repeat that loop again and again.
Exactly. It’s kind of like a debugging mindset.
With AI, your job is not to just define the happy path. You have to find all these edge cases through trial and error.
How did you decide when FigJam AI was good enough to ship?