Process

ProductionAIinfourmoves.

Discover → Conjure → Build → Ship. Eval-first, prompt-first, surface-aware end-to-end. Most teams take quarters — we ship the first version in weeks, with you in the room, not behind a wall of project managers.

The four moves

Discover → Conjure → Build → Ship.

Step · 01

Discover

A week of pairing with you to map the problem, the audience, and the smallest sharpest first version. Honest scoping over heroic estimates.

Typical · 5 working days

What we do

— Pair with your team for ~5 days, in the room or on Zoom
— Audit existing surfaces (product, brand, internal tools)
— Map the actual user flow — not the deck version
— Write the smallest sharpest v1 spec

What you do

— Block calendars for the team that knows the problem
— Open access to product, analytics, and any prior work
— Make the decision-maker available on day 1 and day 5

What gets produced

— Project brief (one page, no fluff)
— Honest scope + timeline + budget
— Smallest sharpest v1 spec
— Decision: continue or stop, no pressure

Step · 02

Conjure

We design in the browser. Static comps are dead — you'll click prototypes within days, not months. Decisions get made on real surfaces.

Typical · 1–3 weeks

What we do

— Design the v1 in the browser (no static comps)
— Build clickable prototypes for every major flow
— Pair daily with you on real surfaces, real copy
— Reject directions that don't earn their place

What you do

— Show up for daily 30-min reviews
— Tell us when the work is wrong; we'll listen
— Provide real content where it matters (no lorem)

What gets produced

— Clickable prototype on a live URL
— Component library in code (Storybook-ready)
— Motion + interaction reference
— Decisions docs for the engineering build

Step · 03

Build

Production build, AI where it earns its keep, observability from day one. Type-safe code, tested paths, shipped behind feature flags.

Typical · 2–8 weeks (scope-dependent)

What we do

— Build production app in TypeScript end-to-end
— Wire AI features with evals + observability
— Set up CI/CD, previews per PR, error tracking
— Pair with your engineering team for handoff

What you do

— Review weekly demos on the staging URL
— Make the call on launch gating
— Ride along on the last 2 weeks of pairing

What gets produced

— Production application, typed + tested
— Eval suite for AI features (golden + regression)
— Observability dashboard + runbook
— Onboarding doc for your in-house team

Step · 04

Ship

Launch with a soft hand. Optional ongoing partnership — weekly cycles, evals on AI features, the unglamorous work of getting better.

Typical · 1 week + post-launch tail

What we do

— Coordinate launch — site, app, comms — together
— Stand by during the first 7 days for incident support
— Hand off the production stack + observability
— Decide together: ongoing partnership or clean exit

What you do

— Pick the launch window
— Take it from here, or invite us to stay

What gets produced

— Live product / brand / surface, observable
— Final handoff doc + access transfers
— Optional ongoing partnership agreement

A week in the studio

What a week actually looks like.

MonStand up

30-min studio sync. What we're shipping this week. What's blocked. What we're saying no to.

TueShip window

First scheduled deploy. Smallest possible change that earns its place in production.

WedBuild day

Heads-down build. Pairing where it earns. Async otherwise. Demo-able by end of day.

ThuReview + decide

Open review with your team. What's working. What isn't. Decisions made on real surfaces.

FriShip window + retro

Second deploy window. End-of-week retro: one thing we learned, one thing we'd do differently.

Tools

The boring infrastructure under the work.

Figma: Visual exploration + handoff to code
Cursor: Code, in the room
Linear: Light-weight project surface
GitHub: Source of truth for shipping
Slack: Async signal
Vercel + Cloudflare: Hosting + edge
Sentry + OTel: Observability + replay
Anthropic, OpenAI, DeepMind: Model upstream

After ship

We hand the work to your team, then either we leave or we stay — but only on terms that earn it.

Most engagements end clean. Some become ongoing partnerships — weekly cycles, evals on AI features, the unglamorous work of getting better. We don't lock you in. Either fit is fine.

Start a project

Walk through it with us.

The Discover week is the most honest call we offer. Tell us what you're making.

Start the conversation