From Brief to Launch in 14 Days: Behind a Real AI-Assisted Build
In April 2026 we shipped a complete custom e-commerce storefront for a Montreal eyewear brand. Brief on a Monday morning, fully live by Friday two weeks later. Custom checkout, 50 product listings, Lighthouse 100 on mobile, French + English. Total elapsed time: 14 days, including 4 weekend days where nobody worked. Here's every day of the build, what AI did, what it didn't, and what this would have cost at a traditional agency.
Day 0: The brief
Monday morning, owner sends a 1-page brief: "We sell eyewear. We have 50 SKUs ready. We need a Shopify storefront with custom checkout, French + English, free shipping over $100, and a B2B wholesale form. Launch as soon as possible." Plus brand reference photos and an existing logo.
We replied with three clarifying questions (color palette preferences, target audience age, any specific competitor sites they like) and a quote: Pro tier scope + e-commerce features, delivered in 2 weeks.
Days 1–3: Architecture + brand system
Day 1 was architecture. Stack picked: Next.js 16 App Router, Shopify Storefront API as headless backend, Tailwind for styling, Vercel for hosting. AI helped with the initial scaffolding — generated the project structure, the basic component library, the Shopify GraphQL queries. About 4 hours of work that would have been ~12 hours without AI.
Days 2–3 were brand. AI generated 6 color palette variations based on the brand brief and the owner's reference photos. Picked the strongest (dark luxury, gold accent, warm whites). Then AI helped draft a complete component library (header, footer, product card, hero, etc.) in the chosen palette. Human spent the time tuning typography (Playfair Display + Inter), refining hover states, and making sure the gold accent had the right weight in dark mode.
AI is great at generating 8 variations of a component in 30 seconds. It's bad at choosing the right one. That's still a human call — taste matters here.
Days 4–7: The build
Days 4–7 were heads-down coding. Product detail pages, collection pages, cart, search, the wholesale form, the contact form, all wired up to Shopify + Resend for email. AI helped with about 60% of the boilerplate — generating component skeletons, suggesting database queries, drafting first-pass TypeScript types.
Where AI fell down: anything involving subtle UX decisions. The variant selector (color + size on the same product) needed careful UX thought — what happens when one color is sold out? How does the gallery update when the user picks a variant? Those weren't code questions, they were product-design questions. Took a human about 3 hours of design + testing.
Days 8–10: Content + photos + product data
Days 8–10 were the unglamorous parts. Uploading 50 product photos to Shopify, writing 50 product descriptions, fixing French translations for every UI string, setting up shipping zones + tax rules, configuring Stripe payments. This is the part agencies usually push to the end and skimp on.
AI helped a LOT with the product descriptions — drafted SEO-friendly first versions for each of the 50 items, which the owner then reviewed and tweaked in about an hour. Without AI, this alone would have been a 2-day task.
Days 11–13: Polish, accessibility, SEO
Days 11–13 were the parts that separate Lighthouse 100 from Lighthouse 70. Every image audited and re-exported (WebP + AVIF, proper sizes for each viewport). Every component checked for accessibility (proper ARIA labels, keyboard navigation, focus states). SEO foundations laid: title tags + meta descriptions per page, sitemap, robots, Schema.org structured data (Product, Organization, BreadcrumbList, Article on blog).
Performance budgets enforced: total JS bundle under 200KB, fonts preloaded, third-party scripts deferred, no layout shift over 0.05. By Day 13 the site was hitting 100/100/100/100 on desktop and 96+ on mobile.
Day 14: Launch + verification
Friday morning Day 14: DNS pointed at Vercel, SSL certificate confirmed, Google Search Console set up, sitemap submitted, analytics installed. Owner ran through the full flow: browse → product → cart → checkout → order confirmation email. Everything green. Site live.
Total elapsed: 14 days. Total developer-hours: ~85. AI-assisted lift: probably saved 60–100 hours vs a fully manual build.
The agency comparison
A traditional Montreal agency would have quoted this scope at $20,000–$35,000 with a 3-month delivery timeline. The owner paid the equivalent of our Pro tier ($499) + Store tier upgrade — well under the agency's rate, in 1/6 the time.
The trade-off isn't quality. The final site outperforms most of the agency-built sites we benchmarked against (we ran Lighthouse on 10 competitor sites in the same niche — none scored above 75 on mobile). The trade-off is that the owner needed to be responsive — answer our clarifying questions within hours, not days. That's the new bottleneck in AI-assisted projects: not developer time, but client iteration speed.
What we used
For anyone curious about the actual tooling:
- Next.js 16 (App Router, Turbopack, Server Components)
- TypeScript strict
- Tailwind + shadcn/ui
- Shopify Storefront API (GraphQL, headless)
- Vercel (hosting, edge functions, image optimization)
- Supabase (newsletter, contact-form data)
- Resend (transactional email)
- Stripe (payments + BNPL via Klarna / Affirm / Afterpay)
- Claude (code generation + content drafting)
- GitHub Copilot (in-editor autocompletion)
Start your project
Brief us in 10 minutes. We'll quote you within 24 hours and (depending on tier) deliver within 1–14 days.
Start your projectRead next
How AI Is Reshaping Web Design for Montreal Businesses in 2026
The real shift isn't "AI built this site." It's that one experienced developer can now ship what used to take a 6-person agency.
What Does a Custom Website Cost in Montreal? (2026 Honest Pricing)
Most agencies hide their pricing. Wix publishes theirs but charges for every basic feature. Here's what a custom site actually costs in Greater Montreal — and what you should walk away from.