The design system built forAI-Human collaboration
While modern teams explore how they might build AI-enabled design systems, I built something more.
DesignerPunk is an ecosystem that pairs a production-ready design system with built-in intelligence layers that guide AI agents to deliver craft at scale.
Why build this system?
When I first tried to build something serious with AI, it became clear: the only way to build something sustainable is to build on something sustainable: a design system.
Challenge
AI enables creation at hilarious speed, but the outcomes of each prompt can be wildly unpredictable — and an agent's motivation to deliver results at any cost can fracture a healthy design and code environment at any time.
On websites, we call this "not ideal"; in Slack channels, we use more four-letter words.
Insight & Thesis
Design systems are shared working agreements between Humans and AI Agents.
- Institutionalized context must persist across sessions
- Every decision must be queryable
- Every decision must be validatable
Goals
- Build a production-grade design system fully optimized for agent-driven development.
- Design for plug-and-play development in a 0-1 product environment — enable small teams to work big.
- Automate workflows for AI-centric collaboration and development processes with humans.
What is this ecosystem?
A mathematical token system that generates native code for iOS, Android, and Web from a single unitless source.
- Tokens carry their own mathematical formulas, validated at build time
- Platform units applied only at final generation — one source, three native outputs
Governance infrastructure of MCP servers and specialized agents for accurate documentation and consistent implementation.
- Rules served via MCP at the point of decision — 82% context reduction
- 8 domain-specialized agents with explicit boundaries and escalation paths
A component architecture that governs cross-platform development through behavioral contracts and a shared API.
- Structured behavioral contracts — inheritable, queryable, with explicit exclusions
- True native implementations per platform, governed by shared contracts
How was this system built?
DesignerPunk was built using Claude models, primarily through Kiro CLI.
What typically takes a team of 15 was built by 1 in under a year utilizing a spec-driven methodology with Human-AI collaboration models.
Every feature starts as a spec. Every spec has requirements, design, and tasks that are formalized and reviewed before a line of code is written. This is how one person maintains consistency in agent-driven development environments.
Beyond the system itself, the most valuable outcome was developing a workflow that makes agent output predictable through the inferable architecture and collaborative systems I architected.
Once the foundations of tokens, components, and frameworks were established, I expanded the scope to explore how I might train agents to build with the system. These explorations included the addition of Google's A2UI specification for agents to build UI for their own native widgets.
Most big-box design systems drive a specific aesthetic. DesignerPunk is built to power yours.
Who is this system built for?
Headcount isn't destiny.
DesignerPunk gives small teams the infrastructure to compete with their largest competition. By making Human-AI collaboration the architecture itself, your team ships quality and scale at the start of the product development journey.
Accessibility from Day One
WCAG 2.1 AA built into every component. Minimize risk and enable +20% of customers you would otherwise exclude.
Three platforms, one source
Web, iOS, and Android tokens and components generated from a single architecture as you need them. Scales to new platforms at your own demand.
Consistency without the overhead
Mathematical token foundations and behavioral contracts maintain quality across every surface, optimized for AI agent development.
Governance that enforces itself
Rules served via MCP at the point of decision. Process without overhead, documentation without hunting.
AI that ships quality
Agents guided by queryable architecture and validatable contracts — structured knowledge, not guesswork.
Code-Design sync
Tokens flow from code to canvas-based tools like Figma, and back to code. One change propagates everywhere.
- Teams of 1–10 who refuse to choose between speed and quality.
- Founders who want Day One infrastructure that scales to Day 1,000.
- Builders who pair human creative direction with AI precision.
- Creators who need opinionated architecture without opinionated design direction.
A small team builds great things. DesignerPunk makes sure they scale.
Humans + Agents of this system
Human
Peter Michaels Allen
Product UX & Systems Architect + Design Engineer
Notable employers and clients
| Employer | Period | Design % | Engineering % |
|---|---|---|---|
| Electronic Arts | Aug – Oct 2010 | 50% | 50% |
| Magi.com | Feb 2011 – Mar 2012 | 100% | 0% |
| eHealth | Apr 2013 – Mar 2015 | 99% | 1% |
| FanTalk | May 2015 – Feb 2016 | 92% | 8% |
| Feb 2017 – Jun 2019 | 100% | 0% | |
| Venmo / PayPal | Jul 2019 – Apr 2024 | 100% | 0% |
| DesignerPunk | 2025 – Present | 45% | 55% |
Agents
Design system agents
Ada
Rosetta System
Token creation, management, and documentation
Lina
Stemma System
Component and contract creation, management, and documentation
Thurgood
Civitas System
Auditing, governance, and spec management
Product system agents
Leonardo
Product planning and tech coordinator
Stacy
Product governance, quality assurance, and applied learnings
Product implementation agents
Sparky
Web implementation lead
Kenya
iOS implementation lead
Data
Android implementation lead
Special thanks to:
- Donnie D'Amato
- Amy Lam
- John Marshall
- Dimitri Otero
- Nicole Fu
- Lauren LoPrete
- Paul Bakaus & Impeccable.style
- TJ Pitre & Console MCP
- Joshua Butner
- Design System Wednesday!
- Tracy Weiss
- The haters and nonbelievers
- Margaret Allen
- Virginia Johnstone (Hi, mom!)
- Kinsley & Colton Kirwan
- Lindsay MacKenzie
Let's build together
For half my career, I've built and managed design systems leveraging my experience in design, research, product management, and front-end development to deliver solutions for Humans — and now AI Agents.
I built this design system with AI agents as my engineering team and primary consumer to meet the needs of this new AI era.
While DesignerPunk's tooling and architecture was built for the purposes of 0-1 to scaffold new products, many of the principles, strategies, and methods are entirely transferable. Imagine what we could build with yours.
If you're interested in discussing more, please reach out to me via LinkedIn.
Problem solve with vision.
Build with collaboration.
Lead by example.