Every designer has gone through a lot of tools in their life. My journey in the UX design field started with something like this: Paper & pencil → Illustrator → Photoshop → PowerPoint (not kidding) → Justinmind → (a long graveyard of other things) → Adobe XD → Figma → and now, amongst many AI tools, Figma Make.

I’ve been using Figma Make for around four or five months—both personally and professionally. This post is about that experience: what I tested, what actually worked, the pricing reality check I didn’t see coming, and the workaround I built when I hit that wall. If you dont know what Figma make is then check these videos below.

Senior Designers Who Opened My Eyes...

When I joined my first UX design consultancy, I met senior designers who were doing wireframing and sometimes even actual visual design in PowerPoint. Not rough wireframes—real, polished UI work, built on meticulously assembled component libraries they’d created themselves. At the time, it was mind-bending. Here I was, coming in with Illustrator skills from my bachelor days, and these folks were producing near-production-quality work in presentation software.

I’ve been thinking about that a lot lately. But more on that at the end.

The Three-Prompt Experiment

This all started with the “tinkering keeda” in me. When I saw the launch of Figma Make, I was intrigued, but I was too occupied by other models and experiments to be an early adopter. It took me some time, but finally, I got to it. Until a few months ago, I had tried Google AI Studio, Claude, Rocket.new, Perplexity, ChatGPT, Lovable.ai, Firebase, Stitch, Replit, and more. So, when the chance came to start experimenting with Figma Make, I needed a controlled experiment to test its capabilities.

I picked a food delivery app—a well-understood problem with familiar patterns and easy-to-judge output quality. I ran the same prompt three times, with different amounts of context attached.

• Experiment 1 — Just the prompt, no references. Figma Make produced something. But honestly, it wasn’t that great. Functional, but generic. Nothing that would make you put down your existing workflow.
• Experiment 2 — Same prompt, plus designs I’d created in Google Stitch. This was noticeably better. The output was much more aligned to what I’d sketched—close to the reference screens, coherent, and clearly influenced by what I’d attached. That felt good.
• Experiment 3 — Same prompt, Google Stitch designs, and a small demo design system I’d built in Figma for this experiment. Clear winner. The output respected the system, stayed consistent, and felt like a designer had been paying attention.

This is where Figma Make earns its reputation. The pattern is obvious in hindsight: Figma Make is as good as the context you give it. A bare prompt produces bare results. Give it reference screens and a design system, and the gap between “AI-generated” and “actually usable” closes in a meaningful way. It genuinely surprised me when I asked it to create multiple variations of one particular screen; some of the elements or concepts it introduced, I had not thought of. This was scary (I had not used Claude until this point).

Picking Your Model (And Why I Used Opus 80% of the Time)

One thing I appreciated about Figma Make: you can choose which model to use, similar to how Rocket.new or Perplexity handles it. There’s a “default/Best” option, but I genuinely couldn’t figure out which model that actually maps to—it’s not clearly documented.

I used Claude Opus for roughly 80% of my experiments. I tried a few other options but didn’t end up using Gemini 3 Flash. The main reason I leaned so heavily on Opus was timing—I started before Figma introduced credit limits. When there’s no ceiling, the natural impulse is to just use it. Sometimes liberally. Sometimes, in retrospect, irresponsibly.

That era is now over.

The AI Credits Wall (And the Gem I Built to Get Around It)

In their March update, Figma started showing how many AI credits each prompt consumes. This was a useful and slightly alarming piece of information. I distinctly remember one prompt that consumed around 2,000 credits in a single shot and another which used 3,000+ credits.

To put that in context: the Figma Enterprise plan gives 4,250 AI credits per user, and the Pro plan gives 3,000 AI credits. One ambitious prompt had eaten nearly half of that or full of that.

So I did what felt natural: I did a “Jugaad” (or built a workaround).

I created a Gemini Gem trained specifically to convert my rough prompts into token-efficient versions. Here’s how I built it: I didn’t know the exact internal logic behind how Figma calculates AI credits, but I knew Figma was routing through models like Claude Opus 4.6. So, I used NotebookLM to research how these models actually handle token consumption—how tokens are counted, how prompt structure affects that count, and which patterns tend to be verbose versus lean. NotebookLM helped me synthesize those resources into a guideline document, which I then used to train the Gem. I did this process twice— But after working on the gem i realised that i can improve this even further and So i built a tool called "promptlens" which can be added as chrome extensio and acts as a live companion app/tool which optimizes our prompts and tells us token usage live. Here is how it looks. i will launch this soon so keep an eye out for my next post.

The Gem usage result: roughly a 60% efficiency improvement. A prompt I would’ve written that needed around 300 tokens, the Gem could convert into something that needed roughly 100. I validated this by using a tool to compare token usage across different Gem frameworks—you could see exactly how many tokens each version was consuming.

if you want a starting point for prompt efficiency in Figma Make, these tools are worth a look, so keep an eye for my next post!

The Real USP of Figma Make — And How Long It’ll Last

Here’s where I land honestly.

Figma Make’s genuine superpower is the depth of its Figma integration. If you’ve built a design system in Figma, you can bring it directly into Make and have the output respect that system Or you can directly edit elements in Figma make ( up to an extent ) as you do in your stitch or Figma file Point and edit the elements specifically. If your whole team is working in the Figma ecosystem, there’s a real collaborative loop available: one person refines the design, another iterates in Make, and things stay connected. That’s valuable.

But the moat is shrinking.

You can already take Figma designs and work with them in Claude. And with Figma’s MCP (Model Context Protocol) integration now available, Claude can interface directly with Figma files. If you already have a Claude subscription, the case for additionally paying for Figma Make credits becomes genuinely hard to make.

The pricing math doesn’t help here. Enterprise gets 4,250 AI credits. Figma sells additional credits at roughly 5,000 for $100. Compare what $100 worth of Claude gives you—in terms of capability, context window, and output quality—and that exchange rate looks increasingly unfavorable. As mentioned in my last blog, even liberal use of Claude opus 4.6 in max plan will not reach usage limits for 80% users.

My honest take: if Figma Make credits come bundled with a plan you’re already on, use them. The tool is capable and the Figma integration is real. But if you’re actively deciding whether to invest specifically in Figma Make tokens versus a Claude subscription, the answer is extremely clear—especially now that Figma MCP exists. The value proposition of Figma Make was always its tight integration with the design environment. That advantage becomes less exclusive by the month.

The value of Figma Make and Figma in itself is reducing even further when Google is coming up with tools like Stitch. It’s just a matter of time before someone creates a link between Stitch and Claude and then—boom! Figma’s moat reduces even more.

The Bigger Thing This Is All Pointing At

Here’s where I circle back to those PowerPoint designers from my first job.

I’ve personally met plenty of people who describe themselves as “UX designers” primarily because they’re fluent in Figma. And I get it—Figma is genuinely powerful, and mastering it takes real effort. But Figma is a tool. If it disappeared tomorrow, the design field and thinking wouldn’t go with it.

The history of product design makes this pattern clear. There were designers who designed products entirely by hand—beautiful, precise, by-hand visualizations shared with management for approval. Then came 3D modeling tools: 3ds Max, AutoCAD, Rhino. And now there are tools which can make 3d products just by prompting!

There Some of the best designers and architects I’ve known couldn’t sketch particularly well. But they could think spatially, and the software gave them a way to express that. The medium changed. The underlying capability didn’t.

Now we’re in the next shift. If you can articulate clearly, you can design. You can build. The layer between your idea and its execution keeps getting thinner. Soon it will be, “If you can think clearly, you can design.”

Some years ago, I used this example in an AI for design workshop I facilitated for design leaders: artists went from hand drawing and painting to Photoshop and Illustrator, and we expanded our definition of who counts as an artist. Then came Midjourney, and it happened again. Someone who was able to articulate visual ideas with precision could now produce images that compete with trained illustrators. Every time the tools change, we rewrite the definition.

We’re rewriting it again now—not just for designers, but for anyone who makes things. The question that matters now isn’t which software you’re fluent in. It’s how clearly you can think, and how precisely you can communicate what you want.

And I keep thinking about those seniors doing wireframes in PowerPoint, with their custom component libraries, making it work with whatever they had. Not because they couldnt use anything else, but because they could utlise anything available to express their thoughts. They weren’t defined by the tool. The tool was just how they showed up that day.

"That is the ultimate lesson. It simply took me an experience with one of the Ai tools to recall a truth I first encountered years ago : the tool is never the architect; it is merely the medium."

---

Tags: AI· Design· Figma· Product· Building in Public