A lot has changed in the world of AI for UI design. I put out an article a couple months ago, and a lot has even changed since then. I wanted to do a deep dive into what has actually stood the test of time in my workflow, what I use consistently every day, and what I'm curious about for the future.

Magicpath is my go-to vibe design tool

For UI design, the big tool that has stood out to me is MagicPath. I have used a lot of different "vibe coding" tools. There's a lot I like about them; some I like more than others. Lovable, I think, is really easy to use. V0 was in my workflow a lot; I liked its versioning and the ability to prototype at a component level. MagicPath takes all of that and really levels it up.

It gives you UI generation, but it provides an open canvas to do it in. You chat to generate stuff, but it lives in a canvas, which is so much better for the way that I work. I want to be able to see things side-by-side, iterate, copy and paste something to save an old version, and generate something new. It's great for that.

MagicPath includes everything I love about these other vibe coding tools. I get version control. I get a design mode where I can make tweaks without prompting. I get live preview links without having to publish a project.

But it does a lot of other things really well, too. MagicPath lets you set styles for your generations, so you can apply specific styles and themes for colors and fonts before you generate something. You can prototype components and then @-mention those components in new generations, which will use them again. This allows you to work in bite-sized chunks and pull those pieces together.

You can also generate variations of an idea. If you have one design, you can generate three different visual styles based on it. Or, if you're running into an issue with behavior or interactivity, you can open it up for the AI to try three different ways to solve that problem. You can generate in flows; if I have one screen, I can generate a new screen that takes into account the style, layout, and context of the previous screen. And, just like any other vibe coding tool, I can open up a preview and use HTML-to-design to bring it back into Figma if I want to make tweaks.

And I can also pull what I'm designing into Cursor. MagicPath provides a terminal command to clone the component and the project you prototyped, and then you can build on that to get a more high-fidelity prototype in Cursor if you want to go in and edit the code and prompt to make changes. That's my main tool. I really, really like MagicPath. It has stood the test of time and is useful to me consistently in real client projects.

Claude is my chat assistant

Claude is still my go-to chat tool. I like its generations for answering basic questions. Claude Projects is my preferred way to handle training a specific chat on context and information, and Claude Research is really helpful for me when I have an open-ended question and need a more specific answer. I've been working on learning more software development, and Claude has been really helpful for that. It can answer my "stupid questions" about running terminal commands or running a dev server on localhost and can answer these very specific questions for me in context.

Granola is my second brain

Granola has become ubiquitous for me for note-taking, it's basically my second brain. It has replaced the meeting recorder I was using before and the paper notes I used to use all the time. With Granola, I can transcribe a meeting and ask questions about it later. I can ask questions about an entire project or all of my meetings. In a lot of ways, it's become a second brain that has been really, really helpful for me in managing my client projects. When I have a lot going on and want to make sure I didn't miss an action item, it honestly feels like I have an executive assistant in the room taking notes and reminding me of things. It's really, really helpful.

Other supporting tools

v0

I used to use V0 more, but I replaced it with MagicPath. I would say of the other vibe coding tools, V0 is the one I like the most. It gives you pretty good control over the design of the code. You can jump between versions really easily and branch off into new chats really easily. You can prototype a component, and it doesn't try to force you to build an entire app. If you're not interested in MagicPath, I do recommend V0; I like it a lot.

Lovable

If V0 is too technical for you, I think Lovable is still the gold standard of the most approachable UI generation tool that's out there. It does all of the main stuff I mentioned: you have a design mode, you have a dev mode where you can access the code. It's very user-friendly, and they've made a lot of improvements since its first launch, where it generates much better UIs. It's really easy to make a prototype and share a link.

Figma Make

I get a lot of people asking me what I think of Figma Make.

The two biggest selling points of Figma Make are that you're already paying for it, and the Figma MCP server is built into it. The Figma MCP server is what allows something like Cursor or Claude to talk to Figma and get specs from your specific Figma frames. It's extremely useful; I highly recommend enabling it and using it in Cursor, or even just in Claude, which can talk to the MCP server. That is where you will get the highest quality design-to-code generation.

If all of that sounds too technical for you, Figma Make effectively has this MCP server built into it. I can paste in a frame and get it built in code pretty quickly. That is true and something that other vibe coding tools that don't integrate with MCP servers can't offer. The other selling point is that you probably already get Figma Make if you or your company are paying for Figma, so that is a worthwhile reason to use it if it's, for all intents and purposes, free.

That being said, it does have a lot of shortcomings. Outside of those two things, I don't think it does anything special. It certainly doesn't do anything better than Lovable as just a baseline. And if it's not recreating your designs, I don't really like its UI generation. I've never seen it generate something that really stood out as effective UI generation; it all has that kind of "AI style" feel with the same repeated colors and gradients. But you can use Figma Make just about the same way that you can Lovable. It's fine; it's just not my favorite tool.

Where I find it useful in my workflow is if I want to recreate a design from Figma, I will generate that in Figma Make, publish it, and then import it to MagicPath with MagicPath's "import from Web" feature. For me, that's most useful if I want to brainstorm with something in MagicPath. It's kind of a workaround to get Figma MCP access in MagicPath. But I typically don't do all of my vibe coding work in Figma Make.

Gemini as an all purpose chat

Like I said, MagicPath is the primary tool I use for UI generation, but I have had good luck generating UIs in Claude, Gemini, or ChatGPT. There are times when I want to see some different approaches or compare different ideas, and I will give the same prompt to multiple tools. Those are typically the cases where I'm using Claude or Gemini for UI generation directly. I will say also, it is nice to have more than one subscription to a chat tool just to have a backup. If for any reason Claude goes down or I hit my usage limits, I know I can fall back on Gemini.

Gemini is also good at image generation, which Claude doesn't offer. And Gemini is another one where its biggest selling point is probably that you either get it through work with your Google account, or it's pretty affordable to upgrade your Google One subscription to the next level to get Gemini access. That's a selling point. It's not my primary tool, but it certainly has its uses, it's pretty affordable, and you can use it in all the ways that I mentioned you use Claude. You can set up "Gems," which are like Gemini's version of Projects. It can do research and web search, so you have access to all of that as well.

Krea for image generation

If you're a designer using AI, you, like me, are probably very sad about the acquisition and the sunsetting of Visual Electric. Visual Electric was what I used to recommend to most designers as the best way to generate images. Kind of like MagicPath, you get an open canvas, which I really, really love. You got access to a lot of models and the ability to upscale resolution or make edits to existing images. It was great. They were acquired by Perplexity and are sunsetting the platform.

What I recommend now as a replacement for it is Krea. Krea is, from my perspective, my favorite alternative to Visual Electric. I certainly miss the open canvas to work in, but for me, I wanted to be able to generate images across a few different models (I use Flux a lot as my default) and generate multiple images at a time, because I don't think it's fair to judge a prompt's quality just based on that one first generation. I want to be able to train custom styles to repeatedly get that same style from other images. I wanted upscaling and editing capabilities. Those are my primary uses. In addition to that, it offers video generation and real-time UI generation where you can draw on a canvas and generate based on a prompt. They have a lot of other editing tools that I haven't even dived into, plus community styles and things. I also like their pricing model. They have pretty affordable subscription plans, and you can buy units of generations without having to upgrade your whole subscription.

I get people asking me a lot, "Why not Midjourney?" For me, I don't work for a company that's going to pay for Midjourney, and realistically, I can't afford to use an image generation tool that is public by default. If I'm going to use this for client work, I need the "secret mode" for Midjourney, and that requires like a $60 a month subscription, which is not realistic for me. So that's my biggest hangup with Midjourney. I do like their image generations, and if the "private by default" feature is not an issue for you, then by all means, that would be a good tool to use. But I like Krea. It's affordable, it does exactly what I need it to, I get access to a lot of models, and I can train custom styles. Those are the big things.

What I'm curious about for the future

Cursor

What I'm most curious about, the big one, is Cursor. I have a lot of opinions about the future of the product design role. I think the future of my role, my business, and the product design role as a whole is that we are going to slowly become design engineers. We will, at minimum, be expected to be able to create code prototypes for the designs we mock up. I don't think we'll be allowed to just live in Figma anymore. To me, that's fantastic. I think that's really exciting.

I have found myself getting frustrated if I have to make a Figma prototype because I get this itch... like I could make this prototype real in the same amount of time using AI, but I'm wasting so much time on the Figma spaghetti of connecting all the prototype links. So what I've been doing is learning Cursor. I feel like that's the next step for me: really getting in the weeds of an AI IDE and AI agents writing code, and really using that as a prototyping tool, at minimum.

I'm really excited about being able to take a MagicPath prototype and build on it in Cursor, and also to use Cursor to explain and help me understand the code and why it's working the way it's working. It's letting me create very realistic prototypes. It's helping me understand the dev side of things better and understand limitations and details that I hadn't thought about with my prototypes. If they were just in Figma, bringing them to life is forcing me to think through them in a lot more detail, which I really like, and kind of solve some problems before handoff.

Where I want to get to is being able to prototype in my product's codebase. To be able to take the products I'm working on, pull the repo, and prototype locally in the actual code base. Someday I'd like to get to the point where I'm submitting PRs and making changes in code. But just in the short term, I am really excited about the idea of making code prototypes, being able to deep dive into it, using the Figma MCP server to bring my designs to life, learning more about React and Tailwind, and thinking of my designs in code. I'm very excited about prototyping things that weren't possible before without code, like interactions that weren't possible, interactions with AI features, and UIs based on real data.

Haiku 4.5 as an affordable coding LLM

As I'm working in Cursor, I'm also very excited about the Claude Haiku 4.5 model. For chat purposes, I'm still going to use Sonnet, but for code writing, Haiku is surprisingly good, and it's a lot cheaper. I like that because I go through a lot of tokens when I'm experimenting with AI and making fixes and changes. It's exciting to me to have a cheaper model to work on that sort of thing.

Flora / Weavy

I'm also curious about some image generation tools like FloraFauna, and Weavy. These are kind of like workflow image generation tools, where you can generate an image and put it through different models to make edits. That was always a world that was kind of above my pay grade; it was never super important for the workflows I was doing, but I was always curious about it.

I'm really curious because Figma has acquired Weavy, so I think it's going to be really interesting to see how that factors into the Figma workspace. In a perfect world, I think Figma could be the real replacement for Visual Electric. So I'm excited for that.

Wrapping up

Those are the big things that stand out for my workflow right now. Some things have stood the test of time, but there's also so much change every single day in the space. Products launch or get acquired or make these huge updates that completely change how they fit into my workflow. It's exciting and kind of unsettling at times.

One thing that really hasn't changed is the importance of being good at prompting and understanding how these models work. As the landscape changes and tools come and go, having an understanding of the underlying models and how to talk to them is still really, really useful.

I'd challenge any designer to spend time trying new tools and giving themselves time to play with these tools and see how they can fit into their work.