I'm glad Figma is pushing in this area. I do think that Figma has one of the best UI datasets out there in the world, and for the most part they've been sitting on it letting it get unused. Excited to see where they take this, but I do think they should be looking at image models in addition to LLM-based edit models (I'm biased as I left figma to start a image-mode-based UI design tool) as I find they're more capable in this area.
Canva designers tend to overcook everything, they lack taste. Easy win for Figma in the long run. That said, the open source Artcraft suite of apps (Figcraft is one of them) have a headless command layer that maps 1:1 to the UI, you can run the entire suite headless on a VPS and use your existing AI subs..
> Artcraft suite of apps (Figcraft is one of them)
This will be pushed to feature and UI/UX parity. The community has spun up an open source Raft/Paxos collaboration server, and are working to connect it with open source cloud and local compute.
Same thing for Canva.
It's not so much that I'm spearheading it. The community is jumping onto this motion quickly. Just look at how well SolidCraft is progressing.
Companies are not going to be able to lock you into tools anymore.
> Enhance 224 to 176. Enhance, stop. Move in, stop. Pull out, track right, stop. Center in, pull back. Stop. Track 45 right. Stop. Center and stop. Enhance 34 to 36. Pan right and pull back. Stop. Enhance 34 to 46. Pull back. Wait a minute, go right, stop. Enhance 57 to 19. Track 45 left. Stop. Enhance 15 to 23. Give me a hard copy right there.
Ed: arguably there's some interesting bits in that scene beyond foreshadowing dejected designers slumped in a recliner, sipping cheap whiskey in front of a screen, adjusting pixel perfect layout across multiple devices using only a rasping monotone voice...
Theoretically that's what the Figma agent is for, but I'll be honest in that it released after I left and I haven't tried it enough to tell whether or not it's effective.
How does training an AI model to understand Figma's editing workflows help it make precise design changes while preserving the original layout and component structure?
I've used various combinations of Codex, Claude Code, GitHub Copilot, Figma's MCP, Figma Agent, Pen.dev, and Paper.design for product design. I have previous experience with front-end dev so I'm more technical than the average product designer.
In my experience, Figma Agent and Figma's MCP leave something to be desired. I'm sure Figma Agent is useful for people who haven't used Claude Code or similar before, but it really requires most of your product's context to exist in Figma for it to do well.
As soon as your product's context exists out of Figma, then it starts to drift and hallucinate its own styles. Using the MCP with an AI that has access to your code provides better results, but even it will have trouble remembering what page I'm asking it to edit.
I know they're trying to address this by adding features such as guidelines and specs, along with linking your Figma files to GitHub repos, but my complaint with these features is a lot of the companies I have worked for do not use the very specific tech stack that Figma supports with these features. If it's anything like code connect, it may take years to get more broad support. Every company I have worked for is a C# Microsoft/Azure house.
This does not help with my broader frustration with Figma which is there is a lack of focus/vision. They're trying to compete on all fronts by developing all these different products/features and it doesn't really feel like they have a cohesive story for what the future or product design is. Granted, I think most of us in this industry don't either, but when am I supposed to use Figma Make vs. Figma Agent vs. Code layers?
They keep blurring the line between them when in all honesty I'm more interested in just using the AI agent I pay for to code what I need and bring it into a canvas for archiving/exploration when I need it.
They should just focus on having a good MCP available with a vision feedback loop (PNG snapshot), good docs, and the frontier models will be able to figure out the rest. Seems ambitious to train your own model. A good MCP really is not hard to make, and it's all they need to do to allow people to use AI well with their product.
Also I'm aware that their Figma specific model is one of many models their agent can use. Even in Figma Make, I find myself getting better results with GitHub Copilot than I do in Make with the same models selected.
Better results meaning it gets closer to what I had imagined in my mind or the design is subjectively better.
I might be wrong or mistaken, but it seems like a big disadvantage for Figma that in order to use an agent with Figma, it has to be a specially trained Figma-aware agent and not just whatever agent I personally want to use.
Feels to me like a big issue that a smart competitor could take advantage of. I want to use my agent, I don't want to pay Figma to use their agent.
I'm also a bit salty about Figma ratcheting down hard on API calls, which creates customer service issues for any product that integrates with the Figma API. So maybe I am biased, but based on comments in here it also seems like the Figma MCP isn't very good either. Is Figma's long-term strategy to go full walled garden?
I've completely replaced my Figma usage by just keeping a `/design/{project}` route where each one mimics a Figma document and I just have the agent build out things in html/css/react/etc
I find it a huge distraction to ever draw anything in Figma now
I've been using design playbooks for years - a page with basically every component laid out using dummy data so you can see reskins all at once. I rock them live in prod, too, not linked anywhere but many of the sites I've built if you go to /design/playbook you get to see the whole thing.
Highly recommend it. I think we should make a .well_known url for them or something, honestly.
Anything would be better than their current embarrassing AI roll out. The only way you can get something halfway decent is to put Claude Code into Fable or some equally high intent window and make an animal sacrifice.
Then the whole thing is effectively useless because of server-side font licensing. Your expensive brand font cannot be used by the MCP so you then need an additional layer of client-side Figma plugins to replace the placeholder font with your real font.
I think figma has numbered months - frontier AI in naked html is getting to the level of design that is honestly stunning - all of those "svg of a penguin riding a bike" training examples must be doing something because I've seen Opus one-shot an SVG of a logo vs scraping a PNG and it was pixel perfect. Claude design, while it came out a bit light, is now very capable.
I know figma has a strong fanbase and companies are all rolling their own "post-trained" models for their built-in AI, and figma is a great product, but the fact frontier models can basically do in html what figma does in it's intermediary format sort of point to where the chips will land in terms of what you will need to use figma for in a few iterations.
I haven't used canva in 6+ months either for the same reason and I don't think either's userbases are growing honestly.
I've been using diffui.ai now for 3 different production grade applications, and haven't looked back. It's missing some features but its ability to actually make designs that look pretty good and plug straight into a coding model is unparalleled in my experience so far.
You've misunderstood the bitter lesson if you think it applies here. It's not "frontier LLMs are always better", it's "algorithms that scale with compute/data will usually end up surpassing algorithms that are based on human understanding".
Fine-tuning an open-source LLM with reinforcement learning in order for it to be better at a specific task does not go against the bitter lesson at all.
> You've misunderstood the bitter lesson if you think it applies here.
That's fair because I am stretching it.
I think there's AI companies that understand and scale AI. Then there's the rest that are doing a modern version of writing domain specific knowledge using skills, and maybe a large collection of their own data.
Maybe these focused models outperform in their own niche similar to how hand-written rules outperformed in their own niches. But long-term, they get outpaced quickly by generalized approaches.
So... I'm not quite sure why I would use this over using GPT/Claude/Gemini directly for design?
Bitter lesson, etc, and training an entire VLM to prevent text overlap seems a bit much, especially since GPT 6.* /Claude 5.*'s vision are good enough that it rarely ever happens for me, and if they do make a mistake, it's generally a 5 minute fix?
I work as a product designer, and your right, mostly people aren't using Figma's AI products (at least in my circles). 80% of my time is spend in Claude Code/Design. 20% in Figma doing manual work when I really need to explore things and spec out in detail.
Figma's current AI Agent is essentially useless, especially when compared to these other tools. Not sure what effect the work in this post will have, but even if it's great I feel that designers have already left Figma and are working elsewhere.
Totally agree with bitter lesson problems here. It’s clear one of the next things taking off right now is the ability to understand and use graphical UIs. I bet they could find some way to get OpenAI to specifically train on Figma, it’s a good use case on both sides. They might do it anyway I guess.
I'm always surprised when my pro-figma sentiment is met with cries of "fanboy!"
Figma has consistently done better than most at nailing the UX. Maybe I'm enough of a generalist that it always feels great to me, and super specialists lament the lack of their super special needs being met (genuinely, a valid gripe for the specialists).
I've tinkered with Figma AI, and it's just good enough to keep me tinkering. It's not creating magic yet, and it's certainly not outpacing claude design to the point where I always pick figma.
I am enthusiastically optimistic that Figma will push the UX on agentic design to where I come to figma every day again.
I'm curious if they'll make this model to be a bit more creative than existing models. I know it's weird to ask for creativity in such case, but I'm already a bit annoyed that Opus 5.5 or Sol or whatever else is quite good at creating clean / consistent but soulless designs. In order to get something really out-of-template it needs a lot of iterations and preps. Or maybe I'm just bad at UX/UI design and my AI setup feels that :D
I'm honestly not sure what took them so long. They've had Figma Make out for about a year/a bit more than a year at this point. Make can call OpenAI, Anthropic, and Google models, and I have no doubt that all those companies structured a deal in a way that lets them view what users are requesting, rejecting, editing, etc and train their models on that information.
With the ability of models to generate design-system aligned interactive prototypes in HTML/Tailwind... there really isn't a purpose for Figma. I use to love it - now I don't touch it. It doesn't really help my process to have not-HTML to work with. Maybe for mobile apps?
Figma, Adobe, Canva all have huuuuge editing model practices that they are licensing to their existing enterprise clients
Its surprising to me that the market is still stuck on SaaSpocalypse not noticing that these companies have pivoted hard in an impressive and revenue generating way
Its to the point where their consumer facing products are a sideshow and distraction, but thats what people are familiar
They’re selling tokens just line all AI forms, with huge enterprise commitment payments up front
>Its surprising to me that the market is still stuck on SaaSpocalypse not noticing that these companies have pivoted hard in an impressive and revenue generating way
The cost of Figmas tokens was a frequent source of consternation at my prior engagement, and this was with a company that gave anybody $750/mo in Cursor credits if they asked nicely and filled out a form. Figma is also facing competition from Claude Design as well as other design tools that have a much cheaper price per seat, and allow you to BYOM. I wouldn't be surprised to see enterprises support a design tool switch if a lead showed how much money the company would save by switching away from Figma to something else.
From being on the other side of the table and seeing many of their clients, I’m seeing clients still pay the upfront engagement.
Claude Design isn’t going to change a corporation’s workflow, an individual contributor isn’t going to cause a workflow change in their own department and especially not another department. Management operates their department in a silo. The outside consultancy driving the executive’s desired AI transformation will. Figma/Adobe/Canva are operating there, and not just coming in with their own software. They just steer things in the direction of also burning tokens.
They’re doing workflow automations not really so vendor locked.
Ex-Figma.
For context, this blog was released the same day Canva made their announcement for Brush, which looks to be further along in this area: https://www.canva.com/newsroom/news/canva-brush/
I'm glad Figma is pushing in this area. I do think that Figma has one of the best UI datasets out there in the world, and for the most part they've been sitting on it letting it get unused. Excited to see where they take this, but I do think they should be looking at image models in addition to LLM-based edit models (I'm biased as I left figma to start a image-mode-based UI design tool) as I find they're more capable in this area.
Canva designers tend to overcook everything, they lack taste. Easy win for Figma in the long run. That said, the open source Artcraft suite of apps (Figcraft is one of them) have a headless command layer that maps 1:1 to the UI, you can run the entire suite headless on a VPS and use your existing AI subs..
> Artcraft suite of apps (Figcraft is one of them)
This will be pushed to feature and UI/UX parity. The community has spun up an open source Raft/Paxos collaboration server, and are working to connect it with open source cloud and local compute.
Same thing for Canva.
It's not so much that I'm spearheading it. The community is jumping onto this motion quickly. Just look at how well SolidCraft is progressing.
Companies are not going to be able to lock you into tools anymore.
Is there a way I can tell figma what to do in natural language?
Bring that layer up, move this box, etc
> Enhance 224 to 176. Enhance, stop. Move in, stop. Pull out, track right, stop. Center in, pull back. Stop. Track 45 right. Stop. Center and stop. Enhance 34 to 36. Pan right and pull back. Stop. Enhance 34 to 46. Pull back. Wait a minute, go right, stop. Enhance 57 to 19. Track 45 left. Stop. Enhance 15 to 23. Give me a hard copy right there.
https://youtu.be/hHwjceFcF2Q?is=GWE74YfHzoEDaiWL
Ed: arguably there's some interesting bits in that scene beyond foreshadowing dejected designers slumped in a recliner, sipping cheap whiskey in front of a screen, adjusting pixel perfect layout across multiple devices using only a rasping monotone voice...
https://blog.kenperlin.com/?p=16063
Install Codex and the Figma plugin.
I haven't edited anything manually in Figma for a while now other than small tweaks.
Theoretically that's what the Figma agent is for, but I'll be honest in that it released after I left and I haven't tried it enough to tell whether or not it's effective.
Had launched this plugin earlier this year to do exactly that: https://www.figma.com/community/plugin/1589249271423490087/b...
it's been literally years and the best figma has to show is an announcement of an upcoming announcement
meanwhile, another commenter was right, the figma mcp is dogshit
How does training an AI model to understand Figma's editing workflows help it make precise design changes while preserving the original layout and component structure?
I've used various combinations of Codex, Claude Code, GitHub Copilot, Figma's MCP, Figma Agent, Pen.dev, and Paper.design for product design. I have previous experience with front-end dev so I'm more technical than the average product designer.
In my experience, Figma Agent and Figma's MCP leave something to be desired. I'm sure Figma Agent is useful for people who haven't used Claude Code or similar before, but it really requires most of your product's context to exist in Figma for it to do well.
As soon as your product's context exists out of Figma, then it starts to drift and hallucinate its own styles. Using the MCP with an AI that has access to your code provides better results, but even it will have trouble remembering what page I'm asking it to edit.
I know they're trying to address this by adding features such as guidelines and specs, along with linking your Figma files to GitHub repos, but my complaint with these features is a lot of the companies I have worked for do not use the very specific tech stack that Figma supports with these features. If it's anything like code connect, it may take years to get more broad support. Every company I have worked for is a C# Microsoft/Azure house.
This does not help with my broader frustration with Figma which is there is a lack of focus/vision. They're trying to compete on all fronts by developing all these different products/features and it doesn't really feel like they have a cohesive story for what the future or product design is. Granted, I think most of us in this industry don't either, but when am I supposed to use Figma Make vs. Figma Agent vs. Code layers?
They keep blurring the line between them when in all honesty I'm more interested in just using the AI agent I pay for to code what I need and bring it into a canvas for archiving/exploration when I need it.
They should just focus on having a good MCP available with a vision feedback loop (PNG snapshot), good docs, and the frontier models will be able to figure out the rest. Seems ambitious to train your own model. A good MCP really is not hard to make, and it's all they need to do to allow people to use AI well with their product.
Also I'm aware that their Figma specific model is one of many models their agent can use. Even in Figma Make, I find myself getting better results with GitHub Copilot than I do in Make with the same models selected.
Better results meaning it gets closer to what I had imagined in my mind or the design is subjectively better.
I might be wrong or mistaken, but it seems like a big disadvantage for Figma that in order to use an agent with Figma, it has to be a specially trained Figma-aware agent and not just whatever agent I personally want to use.
Feels to me like a big issue that a smart competitor could take advantage of. I want to use my agent, I don't want to pay Figma to use their agent.
I'm also a bit salty about Figma ratcheting down hard on API calls, which creates customer service issues for any product that integrates with the Figma API. So maybe I am biased, but based on comments in here it also seems like the Figma MCP isn't very good either. Is Figma's long-term strategy to go full walled garden?
I've completely replaced my Figma usage by just keeping a `/design/{project}` route where each one mimics a Figma document and I just have the agent build out things in html/css/react/etc
I find it a huge distraction to ever draw anything in Figma now
I've been using design playbooks for years - a page with basically every component laid out using dummy data so you can see reskins all at once. I rock them live in prod, too, not linked anywhere but many of the sites I've built if you go to /design/playbook you get to see the whole thing.
Highly recommend it. I think we should make a .well_known url for them or something, honestly.
Anything would be better than their current embarrassing AI roll out. The only way you can get something halfway decent is to put Claude Code into Fable or some equally high intent window and make an animal sacrifice.
Then the whole thing is effectively useless because of server-side font licensing. Your expensive brand font cannot be used by the MCP so you then need an additional layer of client-side Figma plugins to replace the placeholder font with your real font.
> to put Claude Code into Fable or some equally high intent window
What?
I think figma has numbered months - frontier AI in naked html is getting to the level of design that is honestly stunning - all of those "svg of a penguin riding a bike" training examples must be doing something because I've seen Opus one-shot an SVG of a logo vs scraping a PNG and it was pixel perfect. Claude design, while it came out a bit light, is now very capable.
I know figma has a strong fanbase and companies are all rolling their own "post-trained" models for their built-in AI, and figma is a great product, but the fact frontier models can basically do in html what figma does in it's intermediary format sort of point to where the chips will land in terms of what you will need to use figma for in a few iterations.
I haven't used canva in 6+ months either for the same reason and I don't think either's userbases are growing honestly.
The ship has sailed on Figma. It's easier, faster and better to prototype directly via coding agent.
How will Figma outlive Claude Design or any LLM-based design tool in the next 2 years... I really do wonder
I've been using diffui.ai now for 3 different production grade applications, and haven't looked back. It's missing some features but its ability to actually make designs that look pretty good and plug straight into a coding model is unparalleled in my experience so far.
figma is a llm based design tool
Bitter lesson learned in 3... 2..
I know Dylan Field recently posted they made the Llama's 70B model do amazing things but the AI training field is an absolute snake pit.
But then again they have money to burn ig.
You've misunderstood the bitter lesson if you think it applies here. It's not "frontier LLMs are always better", it's "algorithms that scale with compute/data will usually end up surpassing algorithms that are based on human understanding".
Fine-tuning an open-source LLM with reinforcement learning in order for it to be better at a specific task does not go against the bitter lesson at all.
> You've misunderstood the bitter lesson if you think it applies here.
That's fair because I am stretching it.
I think there's AI companies that understand and scale AI. Then there's the rest that are doing a modern version of writing domain specific knowledge using skills, and maybe a large collection of their own data.
Maybe these focused models outperform in their own niche similar to how hand-written rules outperformed in their own niches. But long-term, they get outpaced quickly by generalized approaches.
Ok, so, how is Figma still relevant today when agents can write any design in HTML?
Easier to maintain, scale, and can use most of the components of the design system directly if you're doing web.
So... I'm not quite sure why I would use this over using GPT/Claude/Gemini directly for design?
Bitter lesson, etc, and training an entire VLM to prevent text overlap seems a bit much, especially since GPT 6.* /Claude 5.*'s vision are good enough that it rarely ever happens for me, and if they do make a mistake, it's generally a 5 minute fix?
I work as a product designer, and your right, mostly people aren't using Figma's AI products (at least in my circles). 80% of my time is spend in Claude Code/Design. 20% in Figma doing manual work when I really need to explore things and spec out in detail.
Figma's current AI Agent is essentially useless, especially when compared to these other tools. Not sure what effect the work in this post will have, but even if it's great I feel that designers have already left Figma and are working elsewhere.
> I work as a product designer, and your right, mostly people aren't using Figma's AI products
Yeah, no surprise there.
Totally agree with bitter lesson problems here. It’s clear one of the next things taking off right now is the ability to understand and use graphical UIs. I bet they could find some way to get OpenAI to specifically train on Figma, it’s a good use case on both sides. They might do it anyway I guess.
I'm always surprised when my pro-figma sentiment is met with cries of "fanboy!"
Figma has consistently done better than most at nailing the UX. Maybe I'm enough of a generalist that it always feels great to me, and super specialists lament the lack of their super special needs being met (genuinely, a valid gripe for the specialists).
I've tinkered with Figma AI, and it's just good enough to keep me tinkering. It's not creating magic yet, and it's certainly not outpacing claude design to the point where I always pick figma.
I am enthusiastically optimistic that Figma will push the UX on agentic design to where I come to figma every day again.
I'm curious if they'll make this model to be a bit more creative than existing models. I know it's weird to ask for creativity in such case, but I'm already a bit annoyed that Opus 5.5 or Sol or whatever else is quite good at creating clean / consistent but soulless designs. In order to get something really out-of-template it needs a lot of iterations and preps. Or maybe I'm just bad at UX/UI design and my AI setup feels that :D
There's an unavoidable reason why they create soulless designs
The Figma MCP is so bad, it is almost unusable. Letting the calling AI write some javascript to implement a design?
I moved to OpenPencil after they nerfed the MCP server for free accounts.
Comparing it to sol medium isn’t that great. Opus 5.5 med, or even xhigh is on another level. It works great with pen.dev. They’re doing amazing work!
I'm honestly not sure what took them so long. They've had Figma Make out for about a year/a bit more than a year at this point. Make can call OpenAI, Anthropic, and Google models, and I have no doubt that all those companies structured a deal in a way that lets them view what users are requesting, rejecting, editing, etc and train their models on that information.
With the ability of models to generate design-system aligned interactive prototypes in HTML/Tailwind... there really isn't a purpose for Figma. I use to love it - now I don't touch it. It doesn't really help my process to have not-HTML to work with. Maybe for mobile apps?
Figma, Adobe, Canva all have huuuuge editing model practices that they are licensing to their existing enterprise clients
Its surprising to me that the market is still stuck on SaaSpocalypse not noticing that these companies have pivoted hard in an impressive and revenue generating way
Its to the point where their consumer facing products are a sideshow and distraction, but thats what people are familiar
They’re selling tokens just line all AI forms, with huge enterprise commitment payments up front
>Its surprising to me that the market is still stuck on SaaSpocalypse not noticing that these companies have pivoted hard in an impressive and revenue generating way
The cost of Figmas tokens was a frequent source of consternation at my prior engagement, and this was with a company that gave anybody $750/mo in Cursor credits if they asked nicely and filled out a form. Figma is also facing competition from Claude Design as well as other design tools that have a much cheaper price per seat, and allow you to BYOM. I wouldn't be surprised to see enterprises support a design tool switch if a lead showed how much money the company would save by switching away from Figma to something else.
From being on the other side of the table and seeing many of their clients, I’m seeing clients still pay the upfront engagement.
Claude Design isn’t going to change a corporation’s workflow, an individual contributor isn’t going to cause a workflow change in their own department and especially not another department. Management operates their department in a silo. The outside consultancy driving the executive’s desired AI transformation will. Figma/Adobe/Canva are operating there, and not just coming in with their own software. They just steer things in the direction of also burning tokens.
They’re doing workflow automations not really so vendor locked.
1.) They should have been doing this 3 years ago
2.) You can use any agent with Figma
3.) You can use any agent with not-Figma
I used to really respect Figma's engineering team and have been blown away by the degree to which they've dropped the ball with AI.
All they need is to expose their existing API very well over MCP and give a "screenshot" feature agents can utilize and the agents are good to go.
they need to use nemotron for true sovereignty and stop paying the tax (just like salesforce and servicenow)