Figma Config 2025 was really loud. The keynote talk ran long, the rollout video had twice the views of the previous year, and the AI features had a flair that signaled the company knew it was being measured against an existential threat. I watched the livestream in June with the rest of the design industry, and actually walked away genuinely impressed.
It’s now late October, and the version of Figma open on my desktop is not the version that was on that stage. Some of what was demoed shipped and works. But a lot shipped and is flaky. And some hasn’t shipped at all. But it’s the gap between those buckets, I think, that says more about Figma’s strategy than their noisy Config 2025.
I’m not a Figma insider, and I don’t have product roadmap visibility. What you’re about to read is an outsider’s perspective, only written by someone who opens Figma every day for a paying job and has watched what actually loads in the app over the last four months. If you work at Figma and think I’m getting something wrong, please let me know.
The promises, briefly
For anyone who skipped the keynote, the Config recap post lays out the headline pieces. The shorthand version, in the order they got stage time:
Figma Make was the centerpiece, an AI prompt-to-prototype surface that generates working interactive flows from a text description. The demo showed a designer typing “a settings page for a podcast app” and watching a multi-screen interactive prototype materialize in roughly fifteen seconds.
Figma Sites shipped as a publish-from-Figma website builder, framed as Squarespace plus AI plus the visual fidelity of Figma. The demo showed a designer dragging frames into a stack and getting a deployed website with a custom domain.
Figma Buzz got pitched as the ultimate asset-generation tool for marketing teams, with on-brand image generation and templated layout production at scale.
Figma Draw gave the design surface a real illustration toolkit, the kind of pen-tool maturity that had pushed a generation of illustrators back to Adobe Illustrator.
Figma Grid showed up as a re-imagined autolayout primitive, the answer to a thousand designer complaints about stack-based layout constraints.
A bunch of AI helps got distributed across the existing surfaces. First-draft generation in design files. Smart variable creation. Auto-translation. Auto-cleanup. Auto-connectors in FigJam. You get the idea.
That’s the stuff that was promised. Now, let me walk through what actually landed.
What works the way it was demoed
Figma Draw is the most straightforward win. The vector tools are real and they hold up under sustained illustration work sessions. I’ve watched two illustrators on adjacent teams move their workflow back into Figma since July, after years of doing the design in Figma and the illustration in Illustrator. The seam does appear to be gone. That’s a meaningful product change even though it got the least applause.
Figma Grid is in the product, and it behaves the way the demo suggested. Honestly, it is now my default for any new layout, and it became so quietly. It composes with autolayout, the resizing model is sane, and it plays well with the other features. I haven’t hit a case where I had to back out of grid into something else, which means it was properly thought through.
Smart variable creation works. Translation works. The cleanup helpers work, in the limited sense that they perform the action they promised. Whether you’d actually want them to is a separate question I’ll come back to.
What shipped but is flaky
Figma Make ships. You can use it. It does not produce what the demo produced.
The keynote demo gave the strong impression that you could prompt a complex multi-screen flow and get a clean, branded, interactive prototype out the other end. What you actually get, three months in, is a single screen with reasonable layout, generic styling, and components that don’t always match anything in your design system. Multi-screen flows want a lot of stitching after the generation completes. Brand fidelity is poor by default. Output quality has been inconsistent enough that I’ve stopped using Make for client-visible work and started using it as a sketching tool, which is fine, but it’s not the product that was on that stage.
The interesting bit isn’t that Make is bad. It really isn’t bad. It’s that Make is good in a different way than the demo claimed. The demo claimed it would replace early-stage prototyping. What it actually does is replace a small chunk of the easy stuff at the edges of a design workflow, which is real value, but it’s a much smaller value than the keynote framing suggested.
Figma Sites is similar. It exists, it deploys, and the resulting sites function. They are also, almost without exception, recognizably Figma sites. The visual character is uniform across users, the components are obviously templated, and the SEO and performance story is rough. I’ve heard from a couple of founders who tried it for a landing page and quietly migrated to Framer or Webflow within a month. The product works. The product is not yet competitive in the market it’s selling into.
Figma Buzz is the cleanest example of that gap between what was on stage and what’s in my app. The keynote showed sophisticated brand-conscious image generation. What’s in the product as of last week is closer to a templated layout tool with image slots. The image generation that does exist is generic in a way that makes me think the team hit a quality ceiling they couldn’t move past in time for ship.
What didn’t ship at all
The AI Designer agent, which got a slide and a verbal mention as “coming soon,” has not appeared. The integration between Make and Figma’s existing component libraries (so that Make-generated output uses your team’s actual button instead of a generic one) was demoed and is not in the product. The keynote teased a Code Connect refresh that would push Figma deeper into design-to-code, and what shipped was a maintenance release rather than the rethink that was implied.
This is normal. Every conference keynote oversells the roadmap. Apple does it, Google does it, Microsoft does it, and historically Figma did it less than its peers, which is part of why the misses this year stand out. The Make demo in particular felt like a beat the team needed to hit for narrative reasons, even if the underlying tech wasn’t quite there.
The strategic read
Here’s where the audit gets interesting, and where I want to be careful about what I claim to know.
Figma is hedging, I think. They’re shipping AI features across every surface, but they’re not betting the company on an AI-native workflow. Make and Buzz exist, but they don’t replace the core canvas. The core canvas is still the product. The AI is bolt-on, and the bolt-on is more conservative than the keynote suggested.
I think that’s the right call, given Figma’s position. Their installed base wants Figma to keep being Figma. The risk for them is not that they fail to invent the AI-native canvas. The risk is that someone else does and the canvas stops being where designers spend their day. So Figma is buying time. They are shipping enough AI to look serious to the market, while they figure out what the actual post-canvas product is.
Whether the time-buying works depends on how patient the installed base is. My guess is they have eighteen months before either Cursor, Claude Design, v0, or some not-yet-named product produces the moment when a critical mass of designers stop opening Figma first thing in the morning. Maybe twenty-four. If Figma uses that runway to ship a genuinely re-imagined product rather than incremental AI helps, they keep the category. If they spend it shipping more variations of what’s in the product today, they don’t.
The keynote was Figma signaling to the market. The shipped product is Figma signaling to itself, and the two signals don’t quite line up.
The meta point
That gap between demo and what shipped isn’t just a Figma problem. It’s an industry pattern, and it’s getting worse.
In the last twelve months I have watched Adobe, Microsoft, Google, and Notion all ship keynote demos that materially overpromised what landed in the product. The reason is structural. Generative model output is exciting in a curated demo. The same output is unreliable in the wild. Every product team building on top of these models is making a bet about how much of the demo magic survives contact with real users, and most of them are getting that bet partly wrong.
Why does this matter? Because the buyers are starting to notice. So are the product reviewers. I’ve had two software-purchasing conversations this year where the question “does this actually do what the demo showed” came up unprompted within the first ten minutes. That’s a market shift. The keynote currency is being devalued, and the products that survive the next two years will be the ones that close the gap, not the ones that widen it.
For Figma specifically, the path forward is probably to ship less and ship truer. Make doesn’t need to do everything; it needs to do one thing reliably. Sites doesn’t need to be a Webflow killer; it needs to produce one kind of site that looks great by default. The temptation to keep matching the keynote story will be enormous. I really hope they resist it.
Where I might be wrong
I want to be honest about my blind spots.
I don’t have visibility into Figma’s enterprise telemetry. It’s possible that Make and Sites are doing better than my anecdotal sample suggests, and that the wins are concentrated in customer segments I’m not in. I’d believe that.
I’m also working from a sample that’s heavily product-design biased. Marketing teams, presentation builders, and people using Figma for non-product work might be having a different experience entirely. Buzz in particular might be landing in a market I just don’t see.
And model quality moves fast. The Make of December 2025 will not be the Make of June 2025, and it might not be the Make I’m describing here. If the underlying model gets meaningfully better, some of these complaints reverse themselves overnight.
What I’m confident about is the strategic frame. Figma is hedging, not betting. The next two Configs will tell us whether the hedge worked. If you read the keynote differently or you have data that contradicts what I’m seeing in my own app, I genuinely want to hear it. Please let me know.
I’ll see you in two weeks.
Jameson