Prefer English?
Prefer English? Switch the website language for a better experience.
Figma has integrated several generative AI building blocks directly into its editor: First Draft generates a first interface mockup from a text prompt, while Make Designs (initially announced as Figma Make) takes the concept further by producing interactive prototypes and even usable code. There are also quieter but equally useful day-to-day features: automatic layer renaming, component variant generation, smart content resizing, and layout suggestions based on the most common patterns across the Figma ecosystem.
For an agency like Mazette.co, which runs projects end to end between Figma and Webflow, the question isn't "should we use Figma AI" -- the answer is yes, without hesitation. The real question is: at which stage of the process, on what type of project, and with what human safety net.
First Draft turns a text brief into a visual interface in seconds. On paper, it's spectacular. In practice, it's an excellent rough-drafting tool and a very poor final deliverable tool.
On a recent redesign project for a B2B SaaS client in logistics, the Mazette.co team used First Draft to generate three landing page directions in under ten minutes -- work that, with the usual back-and-forth of moodboards and wireframes, would have taken half a day before the first client presentation. The result: those three directions served as a basis for discussion, none survived as-is beyond the scoping meeting. The proportions were generic, the visual hierarchy flat, and above all none of the proposals respected the client's brand identity, for the simple reason that First Draft doesn't know it.
What First Draft genuinely speeds up:
What it doesn't do: understand the brand, respect an editorial tone, or produce a visual hierarchy designed to convert. That last step remains, and will remain, work handled by humans through product design expertise, people who actually know the client's business.
Make Designs goes further than First Draft by generating interactive interfaces and directly usable frontend code. This is where the feature becomes genuinely interesting for a Webflow-focused agency, because it partially reshuffles the line between design and development.
On an internal dashboard project for an e-commerce client, Mazette.co tested Make Designs to quickly prototype a stock management interface with complex interactions -- dynamic filters, sortable tables, loading states. The generated prototype was functional in about twenty minutes, compared to the several hours of manual prototyping usually needed to test these interactions with the client before validation. The time saved on this specific phase -- testing interactions before design validation -- was real and measurable.
That said, the code generated by Make Designs is not production code. It serves as a proof of concept, not a final development base. On a Webflow project, the challenge remains translating the visual and interactive intent validated in Figma into a clean Webflow structure, with consistent classes, reusable component logic, and careful integration following Webflow best practices -- something neither First Draft nor Make Designs can natively handle.
| Project stage | Real contribution of Figma AI | What stays human |
|---|---|---|
| Scoping / ideation | Quick generation of multiple visual directions | Art direction choices and brand consistency |
| Interaction prototyping | Functional prototype generated in minutes | Real UX validation with target users |
| Design system | Component variant suggestions | Structuring, naming, token governance |
| Handoff to Webflow | Indicative code, not production-ready | Clean integration, classes, responsive, performance |
Figma's AI features increasingly rely on a file's existing components to generate variants or layout suggestions. That's great news when the design system is clean, documented, with well-named tokens. It's a problem when it isn't.
Mazette.co observed this firsthand with a client whose Figma file had accumulated duplicated, poorly named components over several years. Using variant generation features in that context only amplified the mess: the AI proposed variants consistent with the bad practices already in place, reinforcing design system debt rather than resolving it. The lesson is clear: the more powerful Figma AI becomes, the more the quality of the upstream design system becomes a non-negotiable prerequisite, not a cosmetic detail.
This is one more reason why auditing and cleaning up the design system remains a worthwhile step before adopting these tools at scale -- a project Mazette.co systematically recommends ahead of any large-scale Webflow project, through its dedicated methodology.
The most common fear among freelance designers and small studios is simple: if AI generates mockups in seconds, why pay an agency? The reality observed in the field is more nuanced.
The most mechanical tasks -- rough layout drafting, variant generation, quick exploration -- are indeed shrinking in billed time. A creative direction that used to take two days now takes a few hours. But an agency's added value has never come down to the speed of producing a screen. It lies in brand strategy, a deep understanding of the client's business challenges, weighing up options, and the ability to turn a design intention into a product that actually works once it's live -- a bridge Figma AI cannot build on its own between a design file and a high-performing Webflow site.
In practice, this pushes serious agencies toward repositioning: less billing for time spent on pure execution, more value placed on consulting, art direction and end-to-end technical integration. Agencies that only sold mechanical mockup execution will feel the pressure first. Those that sell a brand vision paired with flawless technical execution come out stronger.
Three limitations come up consistently across projects tracked by Mazette.co in 2026, and deserve to be stated clearly before integrating these tools into an agency workflow.
There is also a quieter risk: the temptation to skip user research or competitive analysis because "the AI already suggests something." That is exactly the opposite of what should happen. The faster the generation, the more solid the upstream brief and strategic scoping need to be to properly steer the tool.
{{webdesign}}
At Mazette.co, the position is clear and deliberate: Figma AI has a place in the workflow, but at one specific stage -- rapid exploration and idea prototyping, never final decision-making. The team now systematically uses it during brainstorming sessions with clients, to bring directions to life in meetings rather than describing them out loud. The gain isn't just in production time: it's also in the quality of the dialogue with the client, who can visualize much faster what they like or don't like.
That said, no interface generated by First Draft or Make Designs has ever left a Mazette.co project without going through a full art direction rework, a consistency pass against the existing design system, and then a Webflow integration built for performance and SEO. AI shortens the path to the first idea. It does nothing to shorten the work that turns that idea into a digital product that actually converts.
For agencies still hesitating to build their workflow around these tools, the best entry point remains an audit of the existing Figma setup before anything else -- find best practices on Mazette.co's Figma resources page, or get in touch with the team to put your current design system to the test.
Code
Over 120 companies have already stepped out of the shadows with us.
No, not on the tasks that really matter. Figma AI speeds up visual rough drafting and rapid prototyping, but it understands neither brand strategy nor the client's business challenges. Art direction, creative judgment calls and technical integration remain human work carried out by designers.
No, the code generated by Make Designs serves as a proof of concept to validate interactions, not as a production base. A clean Webflow integration requires reworking the structure, classes and responsive behavior to meet performance and SEO best practices.
Yes, it is even a prerequisite. Variant generation features rely on the existing components in the file: if the design system is poorly structured, the AI amplifies that disorder instead of fixing it. Auditing the Figma file beforehand prevents making an existing design system debt even worse.
Impactful titles, polished meta descriptions, sitemaps, Schema.org, Hn structure, and audits: these are the SEO foundations that make your content readable by both Google AND generative AI. We cover both simultaneously.
With over 180 clients, our agency is one of the few in France to hold the Webflow Premium Partner label, supporting you in every area, including custom web design, sophisticated automation, conversion optimization, and SEO and GEO visibility.
We create bespoke web designs focused on performance, user journeys, and conversion. Join over 180 clients who have trusted us to elevate their brand.
