Three down. Two up.
A May 3 check-in.
April was dense. Some things got smaller. Some things got bigger. Both directions count as a win when they're moving the right way. This is a quick honest read on five of them.
Down 1: the cost of an end-to-end app build
Earlier in April we ran a complete app pipeline test — lead intake through the portal, every milestone from architecture to marketing site, governance agents at every gate, the whole loop end-to-end. Total cost in AI calls: about $22.
The all-model-on-Opus equivalent of the same run would have cost about $60.
We didn't change what gets built. We changed which model handles each step. Faster, cheaper models do quick repeatable work — backend scaffolding, API wiring, integration tests. The slower, more deliberate model is reserved for architecture, client communication, and anything that requires judgment.
What this means for you: we hold prices steady. Lower internal cost on AI-assisted builds means our quotes stay where they are instead of drifting up.
Down 2: the time from "we have an idea" to "the site is live"
The marketing site you're reading right now was a redesign we'd estimated at three sessions of focused work — somewhere between fifteen and twenty hours.
It shipped in half a day.
The thing that compressed it wasn't a magic trick. It was a single batched render of fifteen illustrations that would have taken three to five hours of operator time to produce one at a time. Combined with AI-assisted writing once the vocabulary was locked, the per-component manual estimate beat itself by roughly 3–4×.
The same compression applies to your project. Builds that used to take three calendar weeks now ship in the same window with fewer operator hours behind them.
Down 3: the number of operator clicks per project
A month ago, moving a client project from signed lead to live URL involved roughly a dozen separate operator actions. Domain binding. Server provisioning. Repo creation. Prototype generation. Invoice generation at phase boundaries. Governance review at sign-offs. Each step had its own click.
Most of those clicks are gone. The portal handles them in the background, and surfaces explicit human approval only at the moments that genuinely require a decision — paying an invoice, signing off on a prototype, approving a launch.
The toolmaker thesis: the fewer hands UWC has on the wheel, the more the tool — not the operator — is doing the work. Which is the whole point.
Up 1: public tools
Wave B shipped this month. Two packages went public on npm and GitHub:
- @upstate-web/uwc-skills — three Claude Agent Skills extracted from production client work. A per-tenant AI cost guard. A weighted scoring engine with a quantize clause that prevents a silent Decimal-drift bug. A three-gate client sign-off pipeline.
- @upstate-web/uwc-pipeline-mcp — a stdio MCP server that exposes any directory of agent skills as callable tools.
Both MIT-licensed. Every skill in the bundle shipped to a real client build at least once before it joined. Skills that didn't pass the bar — portable, leakage-free, useful outside our own stack — were dropped, not deferred.
Up 2: real client work in public
Three new case studies are live alongside the earlier portfolio:
- Hale Industrial — an AI workbench for industrial real estate acquisitions. Multi-tenant Django + Astro. Scoring engine with regression tests. Per-tenant daily AI spend cap.
- Rivven Health — a multi-tenant therapy practice platform. Booking, AI assists for session notes and treatment goals, Stripe checkout for invoices.
- Workbench Miami — a class booking and membership system for a co-working studio. Recurring bookings, waitlist promotion, member portal.
Plus eight portfolio demos — fictional businesses across eight industries, each built end-to-end through the pipeline as a proof-of-shape so a prospective client can see what their own site would look like before they sign anything.
What this adds up to
Costs and friction are coming down. Public proof and shipped work are going up. That gap is the thesis: every project teaches the pipeline, and the pipeline teaches the next project.
If you're running a small business and you've been waiting for a builder who hands you tools instead of selling you a retainer — that's us. Tell us what you're trying to do and we'll tell you what it costs and how long it takes.
— Joshua