AI-Enabled Delivery
Our Own AI-Enabled Website Build
How we used ChatGPT, Claude, and Gemini inside a senior-led delivery process to design, build, and ship pixelexact.com.
Conceptual illustration of the workflow — not a screenshot of any AI tool or interface
| Client | Pixel Exact — this website |
|---|---|
| Platform | Next.js marketing website (pixelexact.com) |
| Discipline | Strategy, UX/UI, frontend development, content, SEO, QA |
| AI tools | ChatGPT, Claude Code (VS Code agent), Google Gemini |
| Timeline | 2026-09-16 to present — 19 tracked commits across the first 17 days |
THE CHALLENGE
More Than “We Needed a Website.”
This site had to be a practical test of Pixel Exact's own delivery methodology, not just a brochure. That meant it had to:
- Communicate seven distinct services clearly, without the site turning into an undifferentiated wall of text.
- Deliver senior-grade UX and UI, not a templated layout with the logo swapped in.
- Build a fully responsive, accessible frontend that holds up under real scrutiny, not just a demo.
- Get technical SEO, structured data, and AI-crawler readiness right from the start, not bolted on later.
- Keep the architecture maintainable enough that services, case studies, and insights could keep being added without rework.
- Move quickly, because a studio that sells speed has to be able to demonstrate it on its own site.
What if AI could compress the execution cycle without compromising senior-level design and engineering judgment?
That question shaped every stage of this build. AI wasn't treated as a single tool reached for occasionally — it was integrated across strategy, design, development, content, and QA. Human expertise stayed the control layer throughout: deciding what to build, directing how, reviewing what came back, and approving what actually shipped.
THE WORKFLOW
Six Stages, One Connected Process
Each stage had a clear owner for direction, and a clear point where a senior practitioner reviewed the result before it moved forward.
Strategy & Structure
ChatGPT
Think through positioning, information architecture, and the technical approach before any code existed.
Design & Development
Claude, in VS Code
Implement pages, components, and functionality directly in the repository — not isolated snippets to paste in by hand.
Content
Gemini, ChatGPT
Draft and refine service, case-study, and insights copy for a senior review pass.
Review & Refinement
Senior practitioner
Check every AI-produced change for accuracy, tone, and technical correctness before it moved forward.
QA & Optimization
Claude, in VS Code
Run accessibility, performance, and SEO passes, and fix what those passes turn up.
Launch & Iteration
Git / GitHub
Ship in reviewed, version-controlled increments, then keep refining against real production behavior.
TOOL ROLES
Each Tool Had One Job, Not Every Job
No single tool ran the whole build. Each one was used for what it was actually good at, under the same review process.
ChatGPT
Used early, as a thinking partner rather than a content machine: exploring how to structure the site, weighing technology and architecture options, and stress-testing ideas before anything got built. It didn't make the final call on any of them.
Claude (VS Code Agent)
Worked directly inside this repository — creating pages and components, wiring up navigation and forms, refactoring, fixing bugs, and iterating based on review, rather than producing isolated code snippets that still needed to be integrated by hand.
Google Gemini
Supported content production: drafting and refining service descriptions, case-study and insights copy, and supporting page content, with brand voice and accuracy decided by senior review afterward.
VS Code
The environment all of this happened in — where AI-produced changes were read, tested, and either accepted, corrected, or rejected before moving on.
Git & GitHub
Tracked every change as a discrete, reviewable commit. That discipline mattered more, not less, once AI could propose changes quickly — it's the control layer that makes fast iteration safe to undo.
AI + HUMAN REVIEW LOOP
AI Output Was a Draft, Not a Decision
Nothing AI produced went live because AI produced it. It went live after this loop ran, as many times as it needed to:
A concrete example: when a production build started failing on the hosting platform partway through this project, the failure was diagnosed from build logs, tested against two different hypotheses, and resolved by changing the build strategy — each step tracked as its own commit and verified before the next deploy went out. That's the loop in practice, not in theory.
PROMPTING AS PART OF THE PROCESS
From “Ask AI” to a Repeatable Method
Prompting evolved from an ad hoc question into a structured step in production, governed by a persistent, versioned rules document covering architecture, accessibility, SEO, and performance standards that every AI-assisted change had to satisfy.
Before
Ask AI → Get answer
BY THE NUMBERS
What's Actually Verifiable
Every figure here comes directly from this site's own codebase and commit history — nothing projected or estimated.
page and route handlers make up the site — services, case studies, insights, legal pages, and system routes like the sitemap and llms.txt
services and 3 published case studies and insights entries each, all built and documented through the same workflow
commits across the first 17 days of development, each one reviewed before merging rather than generated and shipped in one pass
commits carry a Claude Code co-authorship trailer — the clearest first-party record of where AI touched implementation directly
named AI crawlers explicitly welcomed in robots.txt, alongside a dedicated llms.txt for AI systems that read it instead of crawling the full site
distinct schema.org structured-data types in use across the site: Organization, WebSite, BreadcrumbList, FAQPage, Service, Article, and CreativeWork
WHY FRONTEND EXPERTISE STILL MATTERS
AI Can Write Code Fast. It Can't Decide If That Code Is Right.
Accessibility and performance work don't happen automatically just because AI is involved — they happen because someone with frontend judgment checks for them. On this build, that meant:
- A sitewide skip-navigation link and a consistent, visible focus style on every interactive element, not just the ones that happened to get attention.
- Motion that respects a visitor's reduced-motion preference by default — scroll-reveal animations and the homepage carousel both check for it before doing anything.
- Semantic landmarks and heading structure on every page, with ARIA used where it adds real information (live regions, expanded/collapsed state) rather than sprinkled on by habit.
- Fonts load through next/font with swap behavior, so text renders immediately instead of staying invisible while a web font downloads.
- Images go through next/image across every page that uses one, for automatic sizing and lazy loading.
- Case studies, insights, and service pages are statically generated at build time, not rendered fresh on every request.
SEO, AEO & AI DISCOVERABILITY
Built to Be Read by Search Engines and AI Systems Alike
Discoverability work ran alongside development rather than after it: structured data (Organization, WebSite, BreadcrumbList, FAQPage, Service, Article, and CreativeWork schema across different page types), a sitemap, a robots.txt that names major AI crawlers explicitly, and a dedicated llms.txt describing the site for AI systems that read it directly. None of this guarantees rankings or visibility — it removes the structural reasons a page would be hard to find or hard to parse.
WHAT WE LEARNED
What This Means for Client Work
AI works best inside a structured workflow, not as a one-off tool pulled out for isolated tasks — this project ran against a persistent, versioned rules document covering architecture, accessibility, SEO, and performance standards, and that context measurably shaped output quality.
Better context produces better output. The difference between a prompt and a brief is the difference between a guess and a decision.
Senior review doesn't slow AI-assisted work down so much as it decides what's actually worth shipping.
Version control matters more, not less, once changes can be proposed quickly — it's what makes fast iteration reversible instead of risky.
AI's value compounds when it's part of the whole delivery process — strategy through QA — instead of bolted onto just one stage of it.
The point isn't that AI means anything can be built instantly. It's that the same workflow — AI acceleration, senior direction, structured review, version-controlled iteration — applies directly to client engagements where speed matters but quality, accessibility, and maintainability can't be the trade-off. That's the process behind How We Deliver , not a separate case made just for this site.
QUESTIONS WE GET
Straight Answers
Did AI design or build this site on its own?
No. Every page, component, and piece of content was directed, reviewed, and approved by a senior practitioner. AI accelerated execution inside that process; it didn't replace the judgment behind it.
Which AI tools were used, and for what?
ChatGPT for early strategy and structural thinking, Claude's coding agent for implementation inside the actual codebase, and Gemini for content drafting and refinement. Git tracked every change underneath all three.
Does this mean Pixel Exact can build any site this fast?
It means this workflow compresses execution time without cutting review, accessibility, or quality checks. Actual timelines still depend on a project's scope and complexity.
WHERE THIS CONNECTS
Built Using the Same Thinking We Apply to Client Work
This build drew directly on UI & design systems thinking for its component structure, frontend & application development practice for its architecture, and the same reusable-component instincts behind our Design Toolkit . For more on how AI fits into delivery without replacing review, see How We Compress Enterprise Delivery Timelines into Weeks, Not Months .
Related service
READY WHEN YOU ARE
Build Faster With an AI-Enabled Delivery Workflow
We combine senior UX, UI, and frontend expertise with AI-assisted delivery to help teams move from idea to production faster, without cutting the review that keeps it shippable.
Talk to Us About Your Next Digital Product