Private · Enter password

Incorrect password

← All work
Agentic Infrastructure · 2026 · In progress

The agentic design system

A five-stage plan to build the context layer that lets AI reason from Aero — not just mimic how it looks. The diagnosis, the sign-off, and the work in progress.

Role Author & Lead
Company SpotOn
Timeline 2026 · In progress
Status Stage 1 active
01 · Diagnosis

Seven years to name the problem

After seven years building Aero, I could finally name the root cause of every adoption failure: the system was never documented in a way anyone — human or AI — could reason from. Not just incomplete. Wrong in kind.

AI tools mimic how things look, not how they're built. They produce output that resembles Aero without understanding its structure, its decisions, or its constraints. A component generated by an AI isn't wrong because it looks different. It's wrong because it doesn't know what it's for.

[ Fill in: The specific incidents or patterns that led you to this diagnosis. What adoption looked like before. What "reasoning from the system" would have prevented. ]

110
Semantic tokens
5
Stage roadmap
MCP Token docs AI infrastructure
02 · Roadmap

The five stages

The plan isn't about adding AI features to the design system. It's about building the infrastructure layer the system always should have had — and making it machine-readable along the way.

01 · Active

Foundation

Structured semantic token documentation. Every token gets a name, a role, a usage rule, and a decision rationale.

02 · Next

Connected

Figma and code linked. A single source of truth where token values flow both directions without manual sync.

03

Unified

Component documentation fused with usage context, design decisions, and real production examples.

04

Observable

System health monitoring. Coverage metrics, adoption signals, drift detection across products.

05

Agentic

AI can query, reason from, and build with the system. MCP layer enables agents to use Aero as infrastructure.

Image · 5-stage phase diagram
Foundation → Connected → Unified → Observable → Agentic
03 · Stage 1 in depth

Building the Foundation layer

Semantic token documentation

[ Fill in: What 110 semantic tokens means in practice. How you structured the documentation — what each token entry contains. The difference between a token catalog and a context layer. ]

Decision rationale

[ Fill in: How you captured the why behind each token. What format you used. How this differs from what existed before. ]

Getting leadership sign-off

[ Fill in: How you made the case internally. What the proposal looked like. What objections you addressed. ]

Image · Token documentation format
Name, role, value, rationale, usage rules
Image · Internal proposal / presentation
The case for the agentic layer
04 · The MCP layer

What "agentic" actually means

The end state isn't an AI that can generate a component that looks like Aero. It's an AI that knows Aero well enough to reason about which component to use, why, and whether the thing it's building is consistent with the system's decisions.

The MCP (Model Context Protocol) layer is the engineering interface that makes this possible — a structured API into the design system that agents can query at runtime.

[ Fill in: What the MCP layer proposal looks like. What an agent can ask it. What you prototyped in the mini Aero dashboard. ]

Image · MCP architecture diagram
How agents query Aero at runtime
05 · What's next

Where this goes from here

[ Fill in: Stage 2 timeline and what it requires. What you're building now. What you need from engineering to move forward. ]

[ Fill in: The broader vision — what Aero looks like when all 5 stages are complete. What changes about how SpotOn builds product. ]

Let's talk

Email paucoro14@gmail.com
Elsewhere LinkedIn →