Interfaces that do not look AI-made.

The agent builds, reviews and polishes interfaces with our own design guide. A person chooses the direction and approves the result.

The problem

“You can tell an AI made this.” “It looks nice in the screenshot, but you cannot use the form with a keyboard.” “Nobody can read grey text on grey.”

It happens to any team that builds screens with AI and no written design criteria: the result comes out fast and looks like everyone else's.

How it is done today

Screens come out with the usual shortcuts: decorative gradients, rounded corners on every element, low contrast and forms nobody has tried with a keyboard. Fixing it by hand, screen by screen, gets left for later.

How we solve it

  1. You ask for the work in one sentence, for example “review this page's design and raise the bar”.
  2. The agent works from our own design guide: layout, typography, colour, motion, accessibility and forms.
  3. With that guide it builds the interface, or reviews and polishes an existing one, without the patterns that give away AI-made work.
  4. If there is a design made in the design tool, it reads it and turns it into code.
  5. If there is not, it proposes design directions and takes them into the design tool so they can be compared.

Result

It is how we work every day: the interfaces we build go through that guide, this website included.

What it takes

  • Your code and your design tool

    Access to the interface's repository and, if you have them, to the design files. The agent reads them and proposes; it does not publish.

  • Your brand and your components

    The colours, typefaces and components you already use. Without them, the agent has nothing to be consistent with.

  • Your rules, in writing

    What is acceptable in your company: accessibility, contrast and keyboard requirements. If they are not written down, we write them with you.

  • A person with design judgement

    Whoever chooses the direction and approves the result. The agent proposes and polishes; the decision is theirs.

How it is built

The base is our own written design guide, which the agent reads before touching an interface. It covers layout, typography, colour, motion, accessibility and forms, and lists the patterns that give away AI-made work so they are not repeated.

The agent works with Figma in both directions: it reads a design and turns it into code, or proposes design directions and takes them into Figma.

The guide is used to build, to review and to polish. The agent does not choose the design direction: a person does, and that person also approves the result.

Frequently asked questions

What does it mean for an interface to “look AI-made”?

That it repeats the usual shortcuts: decorative gradients, rounded corners on every element, low contrast and forms nobody has tried with a keyboard.

Does it work with our designs?

Yes. The agent reads a design made in the design tool and turns it into code. It can also propose directions and take them into that tool.

Who decides the final design?

A person chooses the design direction and approves the result.

Related automations

  • Code review

    Four agents review every change before production. A person decides whether it goes in.

  • Agent-ready website

    An AI agent can fill in the website's contact form. A person always sends it.

Part of AI agent development

The next step

Tell us how you work today and we will tell you where to start.