Dario Miličić
← HomeGet in touch

AI workflow

How I build with AI

I build with Claude Code. It drafts the code and the tests; I own the architecture, review every change, and nothing merges until the tests pass. You get working increments sooner, with more tests and documentation than a hand-typed codebase usually gets.

The loop

Four steps, split between me and the agent

  1. 01

    Rules written down

    Me
    Architecture, patterns and the definition of done go into the repo, where the agent reads them.
    Claude Code
    Works inside those rules on every change.

  2. 02

    Plan before code

    Me
    Approve or change the approach before anything is written.
    Claude Code
    Proposes the plan, the files to touch and the edge cases.

  3. 03

    Build behind a gate

    Me
    Steer in small steps: one screen, one branch, one pull request.
    Claude Code
    Writes the code and the tests, which must pass before a pull request opens.

  4. 04

    Review and ship

    Me
    Read every line, test in real browsers and devices, then merge.
    Claude Code
    Drafts the pull request description and keeps the docs current.

Guardrails

What stays the same with or without AI

  • Only with your OK

    AI touches your code only if you agree, and only with the tools you approve.

  • Every line reviewed

    I read and answer for every change before it merges.

  • Tests gate every change

    Nothing merges with a failing test.

  • Your repo, your process

    The rules and skills live in your repository, so your team keeps them.

Mousike logo

Worked example

Mousike

Mousike is a music event booking platform, in production. I built the Flutter web frontend; the backend is a colleague's Django API. Google Antigravity from February, Claude Code from May.

Mousike dashboard ↗Read the case study

Time to production

An 18-month build, shipped in about six months

Estimated for the same scope without AI
18+ months
From first commit to production, with AI
~6 months
Mousike: the landing page with the catalogue of venues, artists and instruments

AI used

Tools and models, from the commit history

Tools

  • Google Antigravity
  • Claude Code

Models

  • Claude Sonnet 4.6
  • Claude Opus 4.6
  • Claude Opus 4.7
  • Claude Opus 4.8
  • Claude Fable 5
  • Claude Opus 5

Set-up

What is in the repository

CLAUDE.md and ARCHITECTURE.md

The layers, error handling, dependency wiring and component naming, written once so every change follows them.

A custom responsive-design skill

Teaches the agent how the Figma tablet (744) and mobile (390) artboards map onto the app, without moving the existing breakpoints.

One screen, one branch, one pull request

The agent runs the tests before it opens a pull request; I review and merge.

CI deploys every merge

GitHub Actions builds the WebAssembly release and ships it to AWS S3 and CloudFront.

Who does what

The split on this project

Me

  • The architecture and the rules the agent follows
  • Scope, plans and trade-offs
  • Reviewing and merging every pull request
  • Testing in real browsers and on devices
  • Working with the client and the backend engineer

Claude Code

  • Drafting screens from Figma inside those rules
  • Writing unit and widget tests
  • Wiring API endpoints, DTOs and mappers
  • First-pass debugging and refactors
  • Pull request descriptions and commit messages