Figma Make, studied properly: brilliant ideation, brutal credit math

Read time: ~5 minutes.

Prompt-to-prototype tools all promise the same thing: describe it, get it. This week we study the one sitting inside the tool most designers already live in — and the meter running quietly beside it.

📐 The Study: Figma Make

What it actually is. Figma Make turns natural-language prompts into working, code-backed prototypes. You can feed it context — existing Figma frames, design libraries and npm packages via "Make kits," PDFs, even videos — and the output is both visually editable on canvas and inspectable as code. A beta lets you start from your own codebase, and MCP connectors hook it to external tools. That is the pitch in Figma's own words, and mechanically, it holds up.

Where it genuinely helps. Early ideation — and it is honestly good at that. Designer Gregory Tzvi Kaidanov, in a detailed hands-on comparison against Lovable, Vercel's v0 and Bolt, called it "the best AI-driven design ideation tool I've used so far" — fast iteration, working Supabase hookups, and functional prototypes from people who don't write code. His working rule is worth pinning above your desk: use Make for 10–20 rounds of fast ideation, then move into Figma Design for precision and production. It is a sketching tool with a compiler, not a design partner.

Where it falls short. Three limits show up consistently across reviews and Figma's own community forum:

• The outputs are a cul-de-sac. Make projects don't convert into editable Figma Design layers — you get an interactive prototype or a static image, then you rebuild by hand. For a tool living inside Figma, that's the most surprising gap.
• Taste is generic by default. Untethered from a strong design library, results drift toward the same template-flavored layouts every prompt tool produces. Multi-screen flows and information architecture remain the human's job.
• The credit economics are hostile to iteration. A Professional seat includes 3,000 monthly credits, and a long-running Figma forum thread documents how fast they go: one user logged 64 credits to move a toast message 50 pixels; another summarized, "3000 credits is about 45 mins usage." The sharpest complaint is the iteration penalty — many prompts are spent correcting the AI's previous answer, and those corrections bill the same as new work. Costs per action are hard to see until they're gone.

Who should care. Product teams already paying for Figma: yes, for ideation sprints — the design-library grounding is the real differentiator. Non-designer founders: yes, for making an idea demoable before hiring anyone. Solo designers on tight budgets: try it inside the free allowance first, and count your prompts. The tool is good. The meter is the product decision you're actually evaluating.

🗞️ The Digest

• Figma's bigger picture. At Config 2026 in June, Figma announced Code Layers (repos as canvas layers), Figma Motion (a real animation timeline), WebGPU shader fills, generative plugins, and Weave's image tools moved onto the main canvas — Qubika's recap is a solid tour. The direction is unambiguous: design, code and AI in one file.
• Adobe goes agentic. On June 18, Adobe put its Firefly AI Assistant into public beta across Photoshop, Illustrator, InDesign, Premiere and Frame.io, with brand-kit generation, storyboard-to-video, and reusable "Elements" for consistent characters and objects across projects.
• The counter-thesis: canvas as code. Paper, the code-native design tool in open alpha, argues traditional mockups are "beautiful lies" that drift from implementation — its canvas is real HTML/CSS that agents can read and write via MCP. The manifesto is the most interesting design-tool essay of the year so far, whatever you make of the bet.

Have you burned a month of Make credits in a week — or shipped something real with them? Reply and tell us what you built. Honest field reports beat vendor demos every time.

— The Editorial Team