# Judgment, not generation.

AI generates options. I decide which ones ship, and turn those decisions into a system your team can reuse.

## Principles

### AI-enabled, not AI-replaced

AI speeds up the work. Every decision that reaches your product is still made by a person.

### Apps over websites

I design products people use every day, not pages they visit once.

### Systems over one-off screens

Every screen becomes part of a system your team can reuse and extend, with or without me.

### Simplest viable solution

If a simpler approach does the job, that’s the one I recommend. Fewer tools, fewer steps, less to maintain.

### Pragmatic, not hype-driven

New tools earn a place in my workflow when they prove useful in real work, not when they trend.

## From context to system in five steps.

1. **Listen.:** I map users, flows, data and business logic before I design anything. Judgment without context is just opinion.
2. **Define.:** I set the criteria a screen must meet before AI proposes anything. Evaluation becomes a check, not a guess.
3. **Direct.:** I work with whichever tools support an agentic workflow and fit your team. I steer their output toward your product and quality criteria. The tools generate. I decide.
4. **Curate.:** I keep only what holds up and structure it into components and patterns. The system grows, without screens piling up.
5. **Hand Over.:** I deliver infrastructure your team builds on without me. Design, development and product share one reference.

## The pipeline

Figma (Design intent) → Figma MCP (Design context for the model) → Claude Code (Generates the code) → Vue.js (Production components) → GitHub (Review and ship)

AI moves the pixels and the code. I decide where they go.

## NxtPeople: one system, multiple products.

Superflow built NxtPeople-DS for a consistent HR platform and its GoTemp sub-brand.

### The challenge

Multiple HR modules and GoTemp needed one consistent foundation.

### The decision

Shared components, with semantic tokens separating brand from function.

### The rejected option

Separate libraries per product: more maintenance, less consistency.

### The outcome

One system for NxtPeople and GoTemp. New screens reuse shared components.

[Read the NxtPeople case study](/work/nxtpeople)

[Explore all case studies](/work)

## What AI still gets wrong

[TODO: confirm in Bart's words] Honest limits, from daily use:

- [TODO: draft] Consistency: it repeats the patterns it just saw, not the ones your system defines.
- [TODO: draft] States: empty, loading, error and permission states are missing or improvised.
- [TODO: draft] Restraint: it adds options, but rarely knows what to leave out.

## Want this judgment on your product?

[Book a call](https://superflow.design/contact)