H2LooP Brand System (2026)
Nilanjan Debnath
Moving a brand out of my head and into a shared design system that a teammate, a script or an AI agent reads before making anything.
WHAT'S NEW is where the rules live, how they travel, and which jobs stopped needing a designer.
A settled brand with chaotic branding
PROBLEM
The troubles surfaced after H2LooP finalized its brand strategy.
Everyone had their own way of making decks, reports and pages, so nobody could reuse anyone else's document, and quality depended on who made it.
A report sent to one client couldn't be used for another without manually updating 'something' that already looked good, because it looked different from rest of the document.
The decision
Design systems have been solving this for product teams since the Photoshop days. We use Claude throughout all teams, I only needed people to ask people to format their documents before sending it out for it to work org-wide. I chose Claude but it's possible with any AI tool.
Whenever someone creates a collateral, my skills will create sub-agents on their device to do everything in background.
The README opens with an "I want to / Do this" table, from making slides to asking for a change. Written for humans and agents alike
The rules run as code
A pipeline brings any Google Slides export back on-brand: fonts, type scale, system icons, and an overflow check with real font metrics.
build.py regenerates CSS and Tailwind from tokens.json, and a lint check fails any agent skill that drifts.Outcome?
Twelve documents in a day
Six decks and six brochures went from English and Japanese text to branded collateral and were laid out in a day: 1,407 text segments, all from the system's templates.
Anyone can open and reuse anyone's deck, report or page
Engineer or Sales, doesn't matter. Their documents flawlessly come together.
Every shared document meets the same design bar
Team members don't spend time looking up the brand hex code or worry about 'is this looking good?'.
Like this project
Posted Oct 19, 2026
Going from one designer's judgement to a shared database that people, scripts and AI agents all read before they make anything.
