AI broke the model. Not by adding better features, but by making architecture matter more than interface. The question isn't which tool has better AI. It's what the tool is built on, and what AI agents can actually do once they're inside your design files.
We work across both Figma and Penpot at Rangle, shipping production software for teams on tight timelines. Over the past year we've watched AI change what "design-to-code" means in practice. We've seen what works, what breaks, and where the real bottlenecks are. This is what we've learned.
Two models, different tradeoffs
Figma: Platform-controlled AI
Figma is building the AI layer directly into their platform. Code Connect links design components to your codebase. Their MCP server (generally available since mid-2025) feeds design context into Cursor, VS Code, and Claude Code. Figma Make generates working prototypes from prompts. Dev Mode surfaces AI-generated code snippets matched to your actual connected components.
This is real, shipping tooling. Sansan put it into production in August 2025, using Figma's MCP server to generate screen implementations from design components. Not a demo. Production.
The tradeoff is control. What AI can access, how it interprets your designs, what integrations are possible: those decisions live on Figma's roadmap, not yours. For most teams, that's a reasonable deal. Figma is investing heavily in AI and the results are showing. You get a polished, well-supported experience inside an ecosystem that's moving fast. Just know that the ecosystem is Figma's to define.
Penpot: Open ecosystem AI
Penpot made a different architectural bet years ago that's paying off now. Designs in Penpot are expressed as code. Layouts map to CSS concepts. Components mirror web primitives. The design file itself is readable by AI systems without a translation layer.
Their MCP server shipped in December 2025. Earlier-stage than Figma's, but structurally different: it exposes actual design data through an open API and plugin system. Any AI agent, any model, any client can connect. You choose the LLM. You self-host the server. You extend it however you need.
Penpot's AI whitepaper puts it directly: most "AI design" tools follow a "describe, then generate" pattern that demos well but falls apart in production. Their bet is that AI works better when it can read structured design decisions (component relationships, spacing logic, responsive rules) rather than interpreting screenshots or proprietary file formats.
The tradeoff: Penpot's ecosystem is younger. The rendering engine is mid-rewrite, moving from SVG to a Rust/WebAssembly canvas renderer to close the performance gap with Figma. The MCP server is experimental. The community is growing but smaller. You're betting on trajectory over current polish.
What this means for your team
Both approaches work. They point toward different futures. The right call depends on where you are today and where you need to be.
You need a mature, integrated workflow now. Figma's AI ecosystem is further along. Code Connect plus the MCP server gives you a working pipeline from design to implementation today, especially if your team is already deep in Figma's design system tooling. The investment you've already made keeps compounding. That matters.
You need control and flexibility. Penpot's open model means you're not locked into one vendor's AI strategy. Connect any model, host your own infrastructure, build tooling for your specific workflow. For regulated industries or teams with strict data governance requirements, this isn't a nice-to-have. It's a requirement.
You care most about design-to-code fidelity. This is where the architectural difference hits hardest. Penpot's web-standards foundation means designs already speak the language of production code. The gap between "what the designer made" and "what the developer builds" is structurally smaller. Figma's Code Connect bridges that same gap, but it's a bridge you build and maintain yourself.
You're building agentic workflows. Both tools support MCP-based agent integration. The difference is scope. Figma's MCP server exposes design context through their API. Penpot's exposes design structure through an open plugin system where agents can read and write design data directly. More latitude. More risk. More potential.
The performance gap
Figma is faster for large, complex design systems. Period. Years of optimization on a custom C++/WebAssembly rendering engine deliver performance Penpot hasn't matched yet.
Penpot's team knows it. Their rendering rewrite from SVG/DOM to a Rust/WebAssembly canvas pipeline is targeting exactly this gap. Early technical demos look promising, but it's unfinished work. Teams operating at scale should test with their actual files before committing.
What we tell teams
We work across design tools at Rangle. We don't sell either one. When teams ask us which to invest in, we give them an honest read of their situation. Not a brand recommendation.
Most teams with established Figma workflows should stay on Figma. Invest in Code Connect, configure the MCP server, start closing the design-to-code gap within the ecosystem you already have. Figma's AI tooling is strong and getting stronger. And the foundational work you do now (clean component architecture, consistent tokens, disciplined governance) isn't wasted if the landscape shifts later. That work transfers to any tool.
Teams starting new design systems, evaluating data sovereignty requirements, or building AI-native workflows from the ground up should give Penpot serious evaluation. The open-source model, standards-based architecture, and self-hosting option create optionality that's extremely hard to retrofit once you've committed to a closed ecosystem.
Here's the thing that matters regardless of which tool you pick: the design-to-code gap is the actual bottleneck. Not the tool. AI doesn't eliminate the need for clear component architecture, consistent tokens, and disciplined design system governance. It amplifies whatever structure you have. Or whatever mess.
The tool matters less than the foundation underneath it. That's always been true. AI just made it impossible to ignore.
Rangle builds production software in days using agentic development workflows. We work across design tools and technology stacks, helping teams ship faster without accumulating the debt that slows them down later.