up to90%faster
Design-system upkeep and growth
New components, states, tokens and tidying up take minutes, not days.
AI in product design
Design and development are changing fundamentally. AI gives products new possibilities: it saves resources and speeds up familiar stages of work. Fast, high-quality design work is no longer an advantage, it is the baseline. So now is the time to adapt processes to the new tools and keep up with the times.
Proven on a real project
The results are impressive.
AI tools were brought into the work on a live product, and the time for typical tasks was measured before and after. Not a forecast, but numbers from the team’s real work.
up to50%faster
Working out new flows
From task to finished screens twice as fast, and the designer still makes the call.
up to90%faster
Reports and routine
Tasks in the tracker, reports, descriptions and documentation with almost no manual work.
Heavy artillery
Claude + a direct bridge into Figma. AI works inside the files.
The plugin covers the routine around a layout. Big work has its own setup: an AI agent gets the task, reads the product’s rules, writes code for it and runs that code right in the working Figma file. No limits or restrictions of the standard Figma MCP: the agent doesn’t just read the layout, it edits it.
- Task from the terminal, a cloud project or a phone
- Claude reads the product’s rules and writes code for the task
- grids
- spacing scale
- components
- patterns
- quality gates
- Direct bridge a direct channel into Figma, no request limits
- Figma changes in the working file: design system, layouts, states
- Report → decision what was done and what breaks the rules. The designer decides
01
No Figma limits
The standard MCP runs into request limits and mostly just reads the file. The direct bridge runs code in Figma directly, so the agent can change hundreds of layers at once.
02
Code for every task
The setup has no “build a screen” buttons. The agent writes a script for the exact task, so the only limit is what Figma can do, not a tool’s feature list.
03
Rules instead of magic
The intelligence of the process is shared rules, a product profile (grids, scales, components, check gates) and project memory. The agent reads them before every task.
04
Tasks even from a phone
Small fixes go through the terminal, big ones (design-system cleanup, audits, screen generation) go to a cloud project that runs on its own. Both entry points read the same rules.
05
Several products, one process
A new product means a new rules profile, not a rebuilt process. Two design systems with different rules live this way right now, and they don’t mix.
06
Everything under control
A backup before every change, a pilot before bulk changes, design-system changes only after the designer’s “yes”. Rules and scripts live in git with a full history.
A process that keeps improving
No boundaries. The design system always in order.
The way screens are built with AI doesn’t stand still. At first AI assembled new screens only from existing components: the layouts matched the system, but the solutions hit its limits. Now, where it really matters, AI goes beyond the design system, and once a solution is approved it adds the new elements to the system itself.
- Stage 1 Built from the design system
Screens are assembled only from existing components. Correct, but within what already exists.
- Stage 2 The best solution first
If a solution needs an element that’s missing, AI creates it on the system’s styles and scales and marks it as a candidate.
- Now The design system grows itself
After approval, AI adds the new elements to the system on its own: the component, variants, states and description.
This is deep automation of the design process.
And the longer AI works with a project, the better it knows it: every approved decision becomes new knowledge, and the next task starts smarter.
Smart savings
Tokens aren’t infinite. The plugin takes the routine.
To keep costs down, many things are cheaper to do by hand. But even that part of manual work can be automated: the plugin handles most of a designer’s basic actions with a cheaper AI model.
Token spend
Complex tasksClaude + a direct bridge into Figmanew screens, audits, design-system growth
A designer’s basic actionsPlugin on a lighter modelcopy, translation, checks, handoff
Design Assistant — a custom AI plugin for Figma.
The plugin takes over the routine around a layout: it generates components from a description, translates and proofreads copy, checks the layout for mistakes and prepares the file for handoff to development. All in one window, without switching between services and plugins.
Generate from a description
Describe a card or a dialog in words, and the plugin builds a ready auto-layout component on the canvas from 17 block types.
Translation and proofreading
5 languages and spell-checking that keeps styles intact. A stress test shows right away where the layout breaks after localisation.
Linked to the design system
Finds hard-coded colours and fonts and binds them to tokens: one by one or all at once.
Layout report
WCAG AA contrast, overlong copy, unstyled layers and clutter. Click a row to select the layer; much of it is fixed in one click.
Layer hygiene
Removes hidden layers and empty frames and gives clear names instead of “Frame 431”. Everything undoes with a single ⌘Z.
Ready for handoff
A skeleton version of the screen for loading states and an AI spec for the developer right next to the layout.
What’s next In progress
From layout straight to frontend
The next step, already in progress: building frontend straight from the design without a separate frontend developer and shipping changes to the product directly. The foundation is there: precise 1:1 layout against a reference and Figma styles mapped to tokens in code.
Time is money. Delegate the routine to AI and leave the decisions to the team.
Every hour a designer spends renaming states and hunting for broken styles is an hour not invested in making the product better. Now that work can be sped up, and the team can spend the freed time looking for better solutions.