Figma Make Pipeline Pixel Audit

Grant: Boop, boop, boop. Maya: Welcome to UX Autopsy. I'm Grant. Speaker 3: And I'm Maya. All right, Config 2026 is happening right now, at Moscone Center San Francisco. Maya: In-person sold out. According to TechTimes, virtual registration is still free, so no excuses. Speaker 3: No excuses. And the timing on this one is wild, right? Because this is Figma's first major conference as a publicly traded company. And the stock has gone from a 52-week high of $142.92. That's all the way down to around $19. Maya: The math doesn't work, and everyone in that room knows it, which makes tomorrow's Dylan Field keynote and the 2 PM investor session pretty loaded. Speaker 3: Totally. So today we're doing a full autopsy on Figma Make and the MCP pipeline. Does the design-to-production story actually hold up under pressure? Maya: Figma Make hits 60% weekly active usage among their largest enterprise customers by Q1. Impressive number but the question we keep coming back to is impressive where exactly on the canvas or in the actual production code base Speaker 3: Yeah, that's the question. We're going to tear apart the MCP server tools, run a red flag-green flag round on three config features. Maya: and we've got the monday.com engineering team's firsthand account of what happens when CodeConnect isn't wired up spoiler it's not pretty Speaker 3: I mean, I'm going to push back on some of Grant's takes today because I I think there's a real architectural argument here that's worth defending. Maya: You're welcome to try. Speaker 3: Oh, I will. Okay, first segment, let's set the stakes. So, the keynote hasn't even landed yet and I already have opinions. Maya: That tracks. What's the number that's driving all of this? Speaker 3: 60%. TechTimes reported it. Figma Make hit 60% weekly active usage among Figma's largest enterprise customers by Q1 2026. For a tool that launched at Config 2025, that's fast. Maya: It is fast. I'll give them that. But I've been burned enough times by WAU numbers to ask the obvious. Previous follow-up, active how? Speaker 3: What do you mean? Maya: A product manager prompting Make to build a throwaway prototype for a stakeholder deck, that's weekly active usage. An engineer shipping that output to production, completely different animal. The stat doesn't tell you which one. Speaker 3: Okay, that's fair. So you're skeptical before Dylan Field even walks on stage tomorrow. Maya: I'm skeptical as a discipline. Tech Times also says the keynote at Moscone is Wednesday, June 24, 9 a.m. PDT. The expected announcements center on deeper production code base integration and design system intelligence, which, look, if they actually ship that, it answers my question directly. Speaker 3: Right, right. And the reason this keynote has to land is bigger than the product. Maya: The stock. Speaker 3: The stock. Figma peaked at $142.92 right after the IPO last August. TradingView has that on record, and it's been sitting around $19 this week. That's a brutal drop. Maya: Stock analysis reported it's down nearly 40% just this year, so now the same stage that has to convince designers has to convince Wall Street simultaneously. That's a two audience problem I do not envy. Speaker 3: And this is the first major public stage since the IPO, which, I mean, come on, the pressure is real. Maya: If the code base integration announcement is vague, you'll feel it in the price by Thursday. Speaker 3: Totally. Oh, and speaking of pressure, returning listeners will remember our UX debt episode where we basically said Figma's design-to-dev handoff had structural problems. Make was supposed to be the answer. Maya: And there's the uncomfortable part. A 60% WAU number is impressive, but impressive adoption and actually closing the handoff gap are two different claims. Speaker 3: So how do we know if tomorrow's announcement is the real fix or just a... A better dressed version of the same promise. Maya: You look at the architecture, the MCP server, Code Connect, how the token system actually wires into a real code base. That's where the truth lives, not in the keynote slide. Speaker 3: Which raises the question, if the pipeline only works when everything is wired up perfectly, who is it actually built for? Maya: So let's get into the actual plumbing. Three core tools in the MCP server: get design context, get variable defs, and get screenshot. They sound like they work together automatically; they don't. Speaker 3: Wait, what do you mean they don't? Maya: Figma's own documentation, the MCP Server Guide on GitHub, says the model won't use all three unless you explicitly prompt it. Get design context gives you structured React and Tailwind output. Get variable defs pulls your color, spacing, and typography tokens. tokens. Get Screenshot is the visual sanity check. Different jobs, the agent picks one unless you tell it otherwise. Speaker 3: Okay, You're not just pressing a button. Right, and that's where the real travail lives, Maya: because Get Design Context's output, Figma calls it a starting point in the guide, not final code, a starting point. a starting point. Speaker 3: I mean, I actually think that's honest scoping. They're telling you exactly what it is. But it is. Maya: Sure, but if you're the developer on the receiving end of handoff, a starting point sounds very different than it does in a product blog post. Those words don't land the same way. Speaker 3: Fair. So where does it actually break? Maya: Code Connect. That's the critical piece. Code Connect is what maps your Figma components to the actual component files in your code base. If you haven't wired that up-and most teams haven't-the AI guesses. Speaker 3: And by guessing, you mean... Maya: Hard-coded hex values instead of tokens. Brand-new components instead of reusing what's already there. The Monday.com engineering team wrote this up. They tried the naive approach, pasted a Figma link into Cursor, let the MCP run. Their developers still had to refactor the AI-generated code into real components and remove invalid styles. Speaker 3: So you're paying twice, generation time plus refactor time. Maya: Nothing; and the Figma blog actually says it directly: "If those elements aren't mapped via CodeConnect, the server still provides styling context, helping the agent write design informed code from scratch, which is a polite way of saying it's improvising. Speaker 3: Okay, but, and I want to push back here, the architecture is still the right call. You want context fed to the model at query time, not a static export. That's the better model. Maya: I don't disagree on the architecture-the context window approach is smarter than dumping a full Figma JSON file at the top of a prompt. Grant: file at the top of a prompt. But the output quality right now is conditional on setup work most design teams haven't done. CodeConnect wiring, clean layer naming, variables actually applied in the file? Maya: Which is a lot to ask before the keynote is even landed. Grant: Yeah, and on starter plan seats, six MCP tool calls per month, you can burn that in one session. Maya: Wait, six? I could break that debugging a single button. Grant: The math doesn't work for most of the teams this is supposedly for. Dev or full seats on paid plans get proper rate limits, but that's a real cost gate on a product still in beta. Maya: So the practical move for someone listening right now, before Wednesday's keynote, is what though? Grant: Audit your Code Connect coverage before you evaluate Make Codegen. If your components aren't wired, the AI output tells you very little. Little about whether the pipeline works. You're not testing Make, you're testing how well the model can guess. Maya: And that question, whether the keynote actually graduates this from beta infrastructure to something production teams can rely on, that's exactly what design system intelligence and the Figma AI agent are supposed to answer, which we're about to get into. Grant: So pivot from the MCP plumbing. Let's do flags. Figma AI agent. Canvas-native rolled out May 20th. Green or red? Maya: Yellow, honestly. And I know that's cheating. Grant: That's not a flag color, Maya. Maya: I know, I know. Look, the pitch is genuinely good. It lives on the canvas, reads your actual components and tokens, no context switching to a separate sidebar. That part? Green. Grant: Totally green. The no toggle tax thing is real. Every time you leave the file to talk to an AI you lose the thread. Keeping the agent in the canvas is the right call. Maya: But the agent works best when your design system is structured cleanly, per Figma's own blog: mediocre system, mediocre output. Grant: So it doesn't fix the problem we were just describing; it amplifies whatever you already have. Maya: Okay, but that's true of any tool. Grant: Sure, my red flag isn't the concept; it's what happens when the agent picks the wrong component variant. Who catches that before it ships? Maya: Designer catches it? Grant: Right, the human. So the context switching just moved later, not eliminated. Maya: I'll give you that. Still green for me, red flag asterisk. Okay, design system intelligence, auto detecting token inconsistencies, syncing variables, tracking component usage, green or red? Grant: Green on the features, red on the maintenance question. Who owns the rules file the A.I. is reading from? Maya: No one. That's always the answer. Grant: Exactly. Chulbul Designs breakdown of the config announcements flags this, too. The governance piece isn't shipped yet. The detection works, the accountability chain doesn't. Maya: So it'll find your token drift and generate a report and then... sit there? Grant: Until someone schedules a meeting about it, yeah. Maya: Okay, last one: Stigma makes production code base integration. Grant: Red flag, full stop. Maya: Wait, let me at least say it. Grant: It's still in closed beta, Maya. Figma's own blog says that. Tomorrow's keynote is either announcing it's done or it's announcing a rebranding of something half-cooked. Maya: That's wild, right? They're selling a Config ticket around this announcement and the feature isn't out yet. Grant: The math on that is uncomfortable. And I've seen enough product launches where closed beta means we need six more months. Maya: I'm giving it a conditional green. If tomorrow's keynote is a graduation to open beta, I'm in. If it's just a demo, that's your red flag. Grant: We'll know in about 18 hours. But that's exactly the question to take into SEGMENT 4. Not whether individual features look promising on stage, but whether the whole pipeline actually holds up when a real team aims pointed at a real Codebase. So the Monday.com engineering post is kind of the Autopsy photo for this whole conversation. Maya: In the best way. Grant: They pasted a Figma link into Cursor, MCP running, and the code they got back? UI match from a distance. Looked fine, but colors were hard-coded, typography overrode system defaults, manual CSS where design tokens should have been. Their words, from a system perspective, it was a mess. Maya: Right. And the thing they called out specifically? The model wasn't bad, it just had no idea what their design system actually was. Grant: No component names, no valid props, no token references. It guessed. And every guess had to be refactored out by a developer anyway. Maya: So who does this actually serve, Grant? Because that's not a niche team with weird infrastructure. That's most teams. Grant: That's the question I keep coming back to. two. Figma's own blog, The Design Systems and AI piece, calls the MCP a quote productivity coefficient. Only if your design system is already connected through CodeConnect. If you haven't done that wiring, the server still generates code, it's just designed and formed guessing. Maya: Which is exactly my problem with it. The teams that benefit most are the ones who already invested months in Code Connect setup, semantic token naming, component documentation... they were already winning. This just makes them faster. Grant: Okay, but hear me out. Maybe that's not a flaw. Maybe Make is the forcing function. If your team knows bad code Code Connect coverage means the pipeline generates garbage, you finally have a business case to fix the design system debt you've been ignoring for two years. Maya: Or you just have a pipeline that generates convincing garbage faster, which is worse, actually. Grant: I mean, I'm not going to pretend that risk isn't real. Maya: And then there's accessibility. Figma's MCP documentation says WCAG annotations can carry into generated code if designers add them. Nothing in the Make pipeline enforces that. You can have a fully connected Code Connect setup, generate a button component, and still ship without proper keyboard interaction states or focus management. Because... no one added the annotations to the frame. Grant: That's accurate. The Figma dev docs say you can add accessibility notes per component mapping in CodeConnect, but it's opt-in guidance, the generated output still needs a full accessibility review before it goes anywhere in your production. Maya: So the pipeline doesn't respect your design system, it respects however much of your design system you documented for AI consumption. Those are two really different things. Grant: Yeah. And most teams don't know which category they're in until the code comes out wrong. Maya: Wait, so is Make accelerating design system adoption or is it just making the gap more expensive when it shows up? Grant: I don't know. I think both can be true depending. Depending on the team, a mature org with solid CodeConnect coverage, this speeds them up. A mid-sized team with a messy Figma file and no wiring, they're generating a refactor queue, not production code. Maya: And we won't know which direction Figma is actually pushing until tomorrow's keynote. Either Dylan Field shows a real production integration story, actual code base, actual component resolution, or this stays a very polished... demo. Grant: That's the verdict that's still out, and it's the one that actually matters for most of the people listening to this. Maya: So before we get there, there is one thing you can do today regardless of how the keynote lands. Grant: Yep, we'll get into that next. Maya: So, before Dylan Field hits stage Wednesday at 9 a.m., one concrete thing to do tonight. Grant: The rules file. Drop it in your repo root before you touch make. Maya: Yes, Figma's own developer docs spell this out. You tell the AI which design system path to use, which tokens are yours, which components already exist. Without that, it generates against nothing, generic React, hard-coded fills. No idea what's in your code base. Grant: It's the difference between a useful scaffold and a refactor queue. I've seen both. Maya: The refactor queue is... expensive. Grant: Very. So, Maya, scorecard question. We called out in the UX Autopsy episode that mixed production integration was the thing to watch, still closed beta today. Maya: Still closed, yeah. Grant: So tomorrow is the test. Does Field announce general availability with real code-based integration, or does the keynote reframe the timeline and call it progress? Maya: I mean, that's the whole question. Techtimes noted the investor session runs same afternoon at- Known at 2 p.m. PDT, Wall Street's watching the exact same announcement designers are. Grant: No pressure. Maya: Either Mate graduates to production or it extends the demo cycle another quarter. We genuinely don't know yet. Grant: And that's worth watching for. Rules file tonight. Keynote reaction drops after Wednesday. All right, that's the episode. The thing that's sticking with me, Code Connect isn't optional if you want to actually ship real components. Without it, you're just generating expensive throwaway code. Maya: Yeah, and Monday's team proved that the hard way. Refactoring hard-coded colors out of AI-generated output is not a design-to-dev handoff. It's cleanup. Grant: Exactly. And tomorrow's keynote, we're watching one thing. Does production code-based integration ship publicly or does it stay a demo? Maya: Wall Street's watching the same moment at the 2 p.m. investor session, so no pressure, Dylan. Grant: No pressure at all. Maya: Before we go, screenshot a UI that bugs you and tag us. If this episode made you look at your own designs differently, leave us a review. Grant: Subscribe so you catch the keynote breakdown next week. We'll have numbers, receipts, and probably an argument. Maya: Probably. Thanks for listening. We'll see you at the Autopsy.