---
title: AMPECO case study · simbolik
description: An AI design engine built in-house on AMPECO's design system, with an agent that checks its output before a person does. 5× output on design-inclusive tasks, and on-brand work the whole team now makes itself.
project: AMPECO
statement: On-brand work used to wait in one designer's queue. Now the whole team makes it.
meta:
  - Project · 01
  - AI Systems
  - AMPECO
  - In-house
stats:
  - value: 5×
    label: Output on design-inclusive tasks
  - value: 1 → team
    label: On-brand work, from one designer's queue to the whole team
image:
  src: images/work/simbolik-project-full-preview-ampeco.webp
  alt: "The AMPECO logo beside the product's dashboard: charging sessions, energy and revenue, with the CoOperator assistant open beside them, on deep blue."
gallery:
  - src: images/work/ampeco-gallery-image-1.webp
    alt: AMPECO's dashboard with CoOperator, its AI assistant, open beside it, under the line "Everything your platform can do, now through conversation".
  - src: images/work/ampeco-gallery-image-3.webp
    alt: "AMPECO's website, its home page between two more of its pages: \"The AI-native software platform for EV charging operators worldwide\"."
  - src: images/work/ampeco-gallery-image-2.webp
    alt: "Four panels from AMPECO's site: the features for network operators, the AI operations agent, the API and developer tools, and the integrations."
next:
  href: project-designjin.html
service: service-ai-systems
cta:
  title: Does your brand work wait on one person?
  lede: I build the system, the checks and the tools your team makes the work with. Bring one piece that sat in the queue.
  drawing: queue
---

# AMPECO. 5× the output, still on brand.

In my design role at AMPECO, I created the company's new design system and built an AI design engine on top of it. An agent checks its output before a person does. The whole team now makes the on-brand work that used to wait for one designer.

## One queue, *made self-serve.*

### Before

- On-brand assets made by one designer, one request at a time
- The brand's rules were in Figma and in people's heads, where no AI tool could read them

### What I built

- The company's new design system, built in Figma
- The same system written as docs a machine can read, one file per component
- An AI engine on top, not tied to any one AI tool
- A dashboard of self-serve makers, among them a social maker and a diagram maker that exports SVG and PNG, plus an asset library
- An auditor agent that checks every output against the system

## Inside the engine.

Three decisions behind it.

1. ENCODING · **One file per component**

   Every component, style and rule has its own machine-readable file: what it is, where it's used, what not to do. The AI reads the same rules a designer works from.

2. ENGINE · **A base, not an app**

   The engine is a layer under the AI tools. When a better model or tool arrives, the company's design rules move with it.

3. AUDIT · **Its own reviewer**

   An auditor agent reviews each output against the system and flags what drifts, so the designers see work that already passed.

## Where it landed.

It's in daily use. Designers were onboarded, marketing works from it, and teammates have since added features of their own. I was invited onto the company's internal AI board. The design system itself is still being finished, and I've started moving the company's old pages onto it.