Design Systems & AI

AI that neverleaves yourdesign system.

Your team designs and builds with AI agents inside your design system. We install the checks that make that true, in your repositories, verify them against the real screens, and keep them current as tools and models change.

A service led by Sebastián Molinari

Start a conversation

01 — The problem

AI amplifiesinconsistency.

An agent produces in minutes what used to take days. Without constraints, it also produces a new button, a new spacing and a new shade of your brand color every time. Multiply that by everyone on the team and the product drifts in weeks.

Prompts and guidelines do not fix this. Agents do not read guidelines the way people do; they follow whatever actually stops them.

The answer is not less AI. It is a design system the agent cannot step out of.

02 — The method

Four steps,in order.

  1. 01

    Redesign

    Where the product needs it, we redesign first. There is no point in protecting an interface nobody wants to keep.

  2. 02

    Design system

    Tokens, components, patterns and documentation. One source of truth that people and agents share.

  3. 03

    AI guardrails

    Deterministic rules that run on every change: lint rules, checks and the context agents receive before they start. Not suggestions. Checks.

  4. 04

    Supervision

    We install it, verify that it works, and keep watching: new components, new rules, new models.

03 — How the guardrails work

Request,check, result.

Explained for the people who own the product, not only for the people who code it.

  1. Request

    Someone asks

    A designer or developer asks an agent for a new screen, a change, a component. Nothing changes in how they work.

  2. System check

    The system checks

    Before the work is accepted, deterministic rules compare it against your design system: tokens, components, spacing, accessibility. Anything outside is stopped, with the exact reason, before it ships.

  3. Result

    The result fits

    What ships looks and behaves like the rest of your product. Every time, whoever or whatever made it.

04 — What it checks

What the ruleslook at.

  • Colors, type and spacing come from tokens, never from loose values
  • Only approved components are used, and used as designed
  • Layouts follow the system’s grid and breakpoints
  • Accessibility: contrast, focus, labels, semantics
  • Naming and structure of anything new
  • Agent context: what agents are told about the system before they start
  • Visual verification: the result is compared against the design, not only against the rules

Checks are configured for your stack and run in your repositories. Nothing depends on a model behaving well that day.

05 — Engagement formats

Three waysto start.

01

Base

Guardrails on the design system you already have, installed and verified.

The fastest way to stop the drift.

02

Project

Redesign and/or design system, with guardrails installed and verified.

A defined scope with a defined end.

03

Monthly retainer

Ongoing supervision: new rules, new components, agent context kept current, and a direct line to us.

For teams that keep shipping with AI.

Pricing is written per proposal, after a conversation.

You are only as good with AI as you are without it.
Sebastián Molinari20+ years in HTML, CSS, component libraries and design systems. He leads this service.

Bring yourdesign system.We bring the guardrails.

Tell us about your team, your stack and how you are using AI today.