How a project
moves from
sketch to shipped.
Brand work, websites, product builds, AI systems — every engagement is its own shape. The phases below are an outline. The actual route is set by what we're building together.
Five phases.
Conversation.
A long call, then usually a second one. I try to understand the thing, the people behind it, and what would count as success a year from now.
Output · A short brief, written up front, that you sign off on.
Scope.
I map the work into a small number of milestones, each with a price and an output. Nothing is open-ended. If the scope grows, we re-scope together.
Output · A scope document, a price, and a calendar with the dates already booked.
Design.
Most of the design happens at the table with a pencil before it gets to the screen. Reviews are weekly - I talk through the work, you react.
Output · A design system and a set of high-fidelity mockups for every key surface.
Build.
Built statically where I can, and with as little JavaScript as the work needs. You watch it come together on a staging URL from week five onwards.
Output · A working site, app, or system, deployed and ready to take real traffic.
Ship & after.
We launch together, then I stay close for a month, fixing the small things that surface in the wild. After that, the work is yours.
Output · A handover document, training, and a follow-up call thirty days later.
An operating layer,
in three pieces.
When the work calls for it, I build a quiet AI system inside your business — a folder of tools that runs against the work you already have. Each piece is small, named, and readable.
> tools connected. > context loaded. > ready to run.
“Make it run, then make it right, then make it fast — but most software never even gets to ‘run’ properly.”
The room around the model.
Permissions, memory, audit log. The structure that makes AI safe to leave running.
Tools, in plain view.
Each capability is a small program with a clear name. You can read it, edit it, hand it off.
Where your work already lives.
Slack, Notion, Linear, your own database. I meet your tools, not the other way around.
I'm self-taught on Claude Code and work across the frontier models as the job demands - designing agentic workflows end to end, from the prompt design and agent orchestration to the MCP connections into the tools you already run.
JOLI, the studio's flagship build, runs on a multi-stage Claude pipeline and ships live as an integration partner on Hospitable's app marketplace. The day-to-day stack is deliberately small - each tool earning its place against a real job.
Four small jobs that tend to come up.
Triage.
Take the day's incoming email, draft replies in the right voice, and flag the ones that need a human.
Lookup.
Answer customer questions against your archive — past jobs, manuals, that one PDF nobody can find.
Compose.
Turn a Thursday standup transcript into a Friday update for the rest of the company.
Watch.
Quietly check that the thing you said you would do, last week, has actually been done.