This is the first quarterly note from the dispatch desk. The intent of these notes is to be a working diary of what's actually happening — what shipped, what we're working on, what we got wrong, what we learnt. The goal isn't marketing; it's record-keeping.Future cohorts of customers should be able to read back through these notes and see how the practice evolved. So: kettle on, biscuits in the tin, here's where we are.
What shipped
The platform is in production. The generation pipeline now reliably produces a complete, working website from a paragraph of brief, in roughly twenty-five minutes. Forge — the in-product editor that handles changes in plain English — has been running for three months and we've trusted it with our own marketing site. The autonomous-improvement engine is in private beta with a handful of internal demo sites; nothing has gone wrong but we're holding it back from real customers until we've seen another month of clean data.
Hosting infrastructure is solid. Sites are running on enterprise cloud with global delivery network, automatic SSL, daily backups, DDoS protection. Performance budgets — Largest Contentful Paint under 2.0 seconds, Cumulative Layout Shift under 0.1 — are enforced at the publish gate. If a page doesn't hold the budget, it doesn't go live.
What we're working on
The next month is the first-cohort onboarding. Around two dozen customers are queued for the first wave; they're a mixed bag deliberately — florist, plumber, ceramics studio, law firm, coffee roaster, chiropractor, indie game studio, restaurant, gallery. The goal is range. We want to know whether the editorial standard holds across registers we haven't personally designed for, before we open more widely.
The other significant piece of work is site rescue — restoring existing customer websites rather than rebuilding them from scratch. The restoration dossierdocuments the process; the first paid rescue is scheduled for late May. We've quoted six more, four of which are likely to convert. If they all run cleanly, rescue will become a separate department of the desk by Q3.
What we got wrong
Two things worth recording. First, the early version of Forge was too eager to act — it would interpret a vague request (“make the homepage feel warmer”) as licence to redesign half the site. We tightened it. Now it asks for clarification on anything that touches more than one section. Customers prefer the question.
Second, our quality gate was, until February, too forgiving on small interaction issues — hover states that worked on desktop but broke on touch devices, focus rings that were missing on a couple of components. We rewrote that whole layer of the gate; it now catches mobile-interaction issues that production browsers will surface, before publish. Three sites that had passed the old gate had to be re-audited; one needed a small fix; nothing was visible to a customer. Better caught at the gate than caught by a real visitor.
Better caught at the gate than caught by a real visitor.
What the cohort taught us
Three things worth flagging from the first wave of briefs we've read.
One: small businesses don't want websites; they want customers.The site is means, not end. The brief that asks for “a website that works” is almost always asking for “a website that brings me work”. We've been making sure the architecture serves the business goal, not the design impulse, on every brief.
Two: customers think they want fewer pages than they need, then more. The first instinct is “just give me a one-pager”. Then they read through the brief and realise they need a separate page for bookings, and another for the menu, and another for directions. We've been gently pushing back on the one-page request when it's the wrong default; equally pushing back on the eleven-page sprawl when that's the wrong default in the other direction. The architecture is the brief— and the customer often hasn't worked out their own architecture yet.
Three: the “voice” setting in the brief matters more than we expected.Whether a customer ticks “quiet” or “exuberant” or “clinical” or “warm” changes the whole feel of the resulting site — the typography, the spacing, the photography, the body copy. Customers seem to know their own voice better than they think; the briefs that pick a clear register read truer than the briefs that hedge.
That's the desk in May 2026. The next note will be August. If something on the platform breaks badly between now and then, you'll see it surface here too — that's the deal.