How we built winstondigitalmarketing.com.

Context: the pitch is the proof.

Our pitch to prospective clients is that we are AI-native and ship fast. The hardest test case is the one buyers look at first. Our own site. So we did it live, timed it, and kept the receipts.

This page is that receipt. The commit log is real. The time is real. The stack is named. If the claim is exaggerated, you can check.

Two phases. Two stories.

Honest framing: we shipped this in two passes. The first pass produced a 28-page launch-ready marketing site in under 8 hours. The second pass kept stacking scope as new ideas landed. Today the site is 66 pages, 107 commits, and 2 days of end-to-end work.

• Pass one (under 8 hours). Spec, plan, design system, homepage, 12 service pages, About, Our Work index, Playbooks index, Contact, 404, legal. Plus the Markdown-for-bots edge middleware. The site was production-ready when this pass closed.
• Pass two (~34 more hours over the next day). Agentic Workforce flagship service, 10 vertical Playbooks for dental, chiro, plastic surgery, vet, financial, and dispensary verticals. 5 thesis Playbooks from the founder's GEO research. 4 historical blog rewrites preserving GSC traffic. The 8,000-word GEO whitepaper. 11 case studies (7 from Winston's own deck with real metrics, plus 6 from Clutch). The mega menu. NeXT-yellow rebrand. Full entity-graph JSON-LD with Reviews and AggregateRating. GSC-informed redirect map for 49 legacy URLs.

The frame is not "8 hours forever." The frame is "what would have been a 6-week traditional agency engagement, shipped in 2 days." The output exceeded the original scope. Both numbers are honest.

The approach: spec first, then dispatch.

We did not "vibe-code" this site. The method was deliberate and repeatable.

1. Brainstorm. We used the superpowers:brainstorming Skill to interrogate the actual problem. What does an AI-native agency site need that a traditional one does not. Before writing a single line of HTML.
2. Spec. We wrote a design spec the implementation could be checked against. It lives at /docs/superpowers/specs/2026-04-17-winston-site-rebuild-design.md in this repo.
3. Plan. We broke the build into 10 phases with explicit deliverables per phase. Tokens, base CSS, partials, homepage, service pages, content pages, schema, middleware, and so on.
4. Execute. We ran superpowers:subagent-driven-development and dispatched one subagent per phase, in parallel where phases were independent.
5. Iterate. Once the launch-ready site was up, we kept dispatching scoped subagents for each new piece (whitepaper, vertical Playbooks, schema graph, case studies). Same pattern, same voice rules, same review gates.

The first-pass phase-by-phase commit log.

The table below is transcribed from git log --reverse for the first 8-hour pass. Every line is a real commit on a real repo. Total Phase-1-through-Phase-5 execution window: 39 minutes of wall-clock inside that 8-hour project frame.

What we shipped (cumulative, both passes).

• 66 HTML pages across homepage, 13 services, 25 Playbooks, 16 case studies and Receipts, hub pages, legal, and error states
• 65 matching Markdown content twins for AI-bot edge serving
• A Cloudflare Pages Function middleware that detects AI user agents and serves .md with Content-Type: text/markdown
• A complete design system. Tokens, base, components, page styles, partials
• Full JSON-LD entity graph: Organization + ProfessionalService with stable @id, Person with sameAs, WebSite + SearchAction, OfferCatalog of productized packages, BreadcrumbList on every page, Reviews from Clutch testimonials, AggregateRating
• Sitemap, robots.txt welcoming 15 named AI crawlers
• 15 thesis Playbooks (5 from the founder's GEO research, 4 modernized historical rewrites preserving GSC search equity, 6 vertical Playbooks for the verticals that drive traffic)
• 10 more vertical Playbooks for dental, chiropractic, plastic surgery, veterinary, financial advisor, and dispensary practitioners
• An 8,000-word GEO whitepaper at /whitepapers/the-new-rules-of-search/ with 12 jump links and full citation footnotes
• 11 case studies (7 deck-sourced with hard metrics on Winston's own client work, 4 Clutch-imported with public testimonials, plus the Sync Show coordinated-AI case study)
• 62-line GSC-informed redirect map preserving search equity for 49 legacy URLs across ~172K impressions
• NeXT Computer logo-yellow rebrand site-wide with regenerated favicon and OG images
• Mega menu under "What We Do" with 4 columns surfacing every service and AI sub-service directly

Lessons.

Honest ones, not retroactive victory laps.

1. One phase = one subagent kept quality high. A single subagent asked to build the whole site would have degraded noticeably by page 15. Scoping each dispatch to one phase kept each output tight.
2. Parallelizing disjoint phases cut wall-clock meaningfully. Phases 3B + 3C ran together. Phases 5A + 5B (Playbooks and Receipts) ran together. The first-pass 39-minute number depends on this. The second-pass scope additions used the same pattern.
3. The spec was the most valuable artifact. Without a written design doc to check against, the subagents would have drifted on voice, layout, and structure by the third dispatch.
4. The voice rules block was load-bearing. Pasting the banned-clichés list and tone rules into every subagent prompt was the single cheapest intervention that prevented regression to generic-agency voice.
5. The follow-on scope is where the real story lives. Pass one proved we could ship a launch-ready site fast. Pass two proved we could keep stacking high-quality scope on top without degradation. Most agencies cannot do either at this speed. Doing both back-to-back is the actual proof.
6. Real numbers beat hype numbers. We could have written "shipped in minutes." Instead we wrote 39 minutes of execution inside 8 hours of project time inside 2 days end-to-end. Each layer is defensible. Each layer is checkable in the commit log. Hype is vapor. Receipts compound.