We build AI into your product's interface

Command palette, inline generation, web agents — placed at the step the user is already working in.

You might be stuck here

Whether AI works is often not a model problem. It's that the AI sits somewhere the user never goes.

  • The feature exists, nobody finds it

    • The menu keeps growing; things end up three levels deep.
    • Users can't remember which page anything is on.
    • Finished features go unused.

    How we handle itStart from where users actually click, and which menus nobody ever opens.

  • AI is off in the corner

    • A floating bubble you have to open first.
    • You get an answer, then copy it back into the form.
    • Two extra actions and people stop using it.

    How we handle itPut AI in the path of that step, so the result lands straight in the field.

  • No idea what it's doing

    • You press the button and get a spinner.
    • It fails without saying why.
    • Nothing can be undone.

    How we handle itWaiting, failure, uncertainty and undo are all designed in.

Three patterns

All three put AI in a path the user is already following. Start with the one that helps most; you don't need all three.

  • Command palette

    • One input that navigates, runs actions and answers questions.
    • Type and you're there — no remembering which menu it's under.
    • Fully keyboard-driven.
  • Inline AI

    • Generate and rewrite in place, inside editors, forms and tables.
    • The result lands in the field — no copy and paste.
    • Every step is undoable.
  • Web agent

    • “Merge these three orders and send the notification” — done across pages.
    • Every action is listed for approval first.
    • You can stop it halfway.

How one feature goes from idea to live

  1. 01

    Watch the flow

    1 week
    • Watch how users actually work.
    • Find the steps they get stuck on most.
    • Decide which step AI belongs in.
  2. 02

    Design

    2–3 weeks
    • Put AI into the path of those steps.
    • Waiting, failure, uncertainty and undo designed together.
    • Specs and a component library delivered.
  3. 03

    Build

    Scope-dependent
    • Frontend build and backend integration.
    • Responsive layout and keyboard support included.
    • Source code delivered to you.
  4. 04

    Measure and tune

    After launch
    • Look at real usage data.
    • Find what still isn't being used.
    • Change it and measure again.

See the patterns first

The same interaction patterns are open-sourced as dotdotduck: command palette, web agent, inline AI you trigger by selecting text. No install, no signup — open it and try.

See dotdotduck

Pricing

UX / UI package

NT$40k–280k

Usually includes:From an experience base for key flows to the full product UX/UI, including login, permissions and data-structure flow design.

Affects the quote:Number of screens and flows, whether a component library is included.

Final price is confirmed once the scope is clear · First consultation is free

AI-enabled website

From NT$100k

Usually includes:Semi-custom site or modernisation, including one or two AI frontend features and responsive layout.

Affects the quote:Page count, whether it connects to existing systems, whether content needs rewriting.

Final price is confirmed once the scope is clear · First consultation is free

Frontend AI experience

NT$80k–400k

Usually includes:From a single flagship feature (command palette / inline AI / voice input) to multi-feature integration and web agents.

Affects the quote:Number of features, how much backend it connects to, how autonomous the agent is.

Final price is confirmed once the scope is clear · First consultation is free

Common questions

Can you do design only, without the build?

Yes. The UX/UI package is design and specs only, handed to your engineering team to build.

We already have a site. Do we have to rebuild it?

Usually not. A command palette or inline AI can be added to your existing frontend, which is a far smaller job.

What happens on screen when the AI gets it wrong?

That's part of the design. Waiting, failure, undo and “the AI isn't sure” are all designed alongside the happy path.

Can we see something you've built?

Yes. The same interaction patterns are open-sourced as dotdotduck — open it and use it.