// not a case study — this is my AI practice, pointed at itself

01—Why Foundry exists

I design with craft, then I build it in code. Foundry is that habit turned into a place. It started as three separate itches — teach what I'd learned, fix a problem that was slowing me down, and understand the infrastructure I was betting my craft on — and became one roof at withfoundry.ai.

// The front door — withfoundry.ai

02—Teach: AI as a design material

Most designers I know sit somewhere between curious and overwhelmed. The tools change weekly, the advice contradicts itself, and almost none of it is written for people whose first language is craft. So I wrote the resource I wished existed: how working with AI actually benefits a designer, how to get started without drowning, and how to master it as a design material — across every major platform, not locked to one.

That last part is deliberate. The platforms will keep leapfrogging each other. Taste and ways of working transfer; tool-specific tricks don't.

03—Solve: carry context between platforms

Everything moves fast, so I jump. New model drops, new skill ships, new agent worth trying — I want to be there the week it lands, not the quarter after. But every jump charged the same tax: re-explaining the architecture, the decisions, the conventions of whatever I was building. Again. To a different machine.

Foundry MCP is the fix I built for myself. One install, and your project context travels with you — decisions, conventions, and component knowledge stored once and retrieved by semantic search in every agent session, on every platform. It also scans for security vulnerabilities as you go — missing auth, hardcoded secrets — and the findings flow back into memory, so the next session already knows.

04—Understand: build the infrastructure yourself

Opinions about AI are cheap. I wanted mine to be load-bearing — so I built the stack under the material: a memory engine on pgvector semantic search, a REST API, a TypeScript SDK, the MCP server, and the documentation at docs.withfoundry.ai. Designing AI experiences lands differently once you've stood up the retrieval that feeds them.

// The docs — API, SDK, MCP

05—What building it taught me

The context problem is a design problem wearing an infrastructure costume. What to remember, what to surface, when to stay quiet — those are judgment calls, the same taste that decides what belongs on a screen. Building Foundry didn't make me more of an engineer. It made the design opinions sharper, because now they've shipped.