Skip to content

Everything We Built

Goose Digital
ai-agentsautomationclaude
Pixel art terminal window titled week 14 — Everything We Built — with two geese flying over a moonlit city skyline and river, past glowing billboards reading automate scale elevate, AI plus people better together, ship it and good code

Fourteen weeks

This is the last one. Both tools are built, tested, and in the hands of the people who will be using them after we’re gone. So this week we’re doing something different — a full overview of what both tools actually became, from the first line of code to where they’re sitting right now.

Not a retrospective. Just: here’s what we built.

WebAI — from brief to live site

WebAI exists because manual WordPress builds are slow to start, inconsistent across clients, and expensive to maintain. The idea was to replace that process with something that could take a guided intake and produce a complete, production-ready site — automatically, with real brand identity, real content, and a quality bar that doesn’t require a developer to enforce.

The process starts with a seven-step wizard. About fifteen minutes. If the client already has a site, brand colours, fonts, copy, and page structure get pulled in automatically. Nothing is silently assumed — everything gets confirmed before anything builds. The output of the wizard is a single locked BuildBrief JSON that every downstream step reads from. Nothing is improvised after that point.

Claude proposes the sitemap, generates a full design token system for the brand — colours, typography, spacing — and then streams each page in around thirty seconds. A five-page site is done in three to four minutes. The AI cost for a full build sits between ten and thirty cents. Every page comes with a live preview, a decision log, and plain-English change requests. You can regenerate one page without touching anything else.

Before anything ships, an automated Lighthouse audit runs against four categories — performance, accessibility, SEO, best practices. The target is ninety-plus across all of them. That’s the default quality bar, not an optional extra.

CMS is built in. Clients log into an admin page, edit content through simple forms, and changes go live within minutes. Decap handles the lightweight version. Sanity is the upgrade path for content-heavy sites. Deployment is one click — GitHub repo, Cloudflare Pages, live.

The account lead runs the intake. The developer receives a locked package with everything documented. Every brief is saved and resumable. That’s the whole handoff.

Image AI — from brief to finished visual

Image AI is an AI design studio. You describe what you need, pick a brand, and the tool composes a finished, on-brand marketing visual — layout, imagery, copy placement, brand consistency — without leaving the workspace.

It starts with the brand kit. Every brand has a profile — colours, fonts, logos, visual identity styles, assets organised into folders. That’s the single source of truth, and everything the tool generates inherits from it automatically. Logos auto-populate a protected category. Visual identity styles are reusable creative directions that apply at any size. Custom size presets come with safe-zone margins per channel.

To create something, you write a plain-language brief. The AI plans the layout and generates every element. Editable mode keeps every layer live and adjustable on the canvas. Fully baked renders everything into one polished image. Reference images can be attached as components, placed like logos, or used as style references only. Resolution and model selection are both controllable — or left on automatic.

The canvas editor is a full direct-manipulation workspace. Move, resize, rotate, multi-select, align with snapping guides. The object inspector gives full control over text, image, shape, and icon properties. Everything stays editable.

AI editing works two ways. The Magic Wand reads the whole design and changes only what you ask — placement stays, context is preserved, template notes are respected. Isolated Edit is surgical: just one image and your words, best for precise fixes.

The rest of the system is built around reuse:

  • Templates save a canvas as a reusable starting point with fill-in slots and ground-truth notes the AI follows every time.
  • Decompose goes the other direction — it reads an existing design and breaks it into editable layers, making it faster to migrate old templates into the tool.
  • Carousels take one brief and return a planned multi-slide narrative with a shared visual system locking design across every slide.
  • Reflow takes a finished design and recomposes it for a new aspect ratio — text repositions, the hero image outpaints to fit.

Multiple sizes live as tabs in the same project. Channel previews show what the graphic looks like as a real social media mockup before anything gets exported.

Export options cover PNG, JPEG, SVG, WebP, HTML5, and print-ready PDF, with size downsize and image compression built into the export panel. The Prompt Genie is an AI assistant inside the tool that writes, refines, and reverse-engineers prompts. The admin panel lets the team tune models and system prompts without touching the codebase.

What both tools share

Underneath both pipelines is the same idea: brand context should live in the system, load automatically, and shape every output before the user types a single word. The brand intelligence framework that both tools pull from is what makes the outputs consistent, on-brand, and useful — not clever prompting, not manual setup before every generation, but a foundation that’s already there.

WebAI takes a client from intake to a live, deployed website. Image AI takes a brand brief to a finished visual. Both without the manual production layer that used to sit in between.

That’s what fourteen weeks built.