How we use AI and Lovable.dev to build the howtech.uk website
When we tell SMEs that AI can genuinely accelerate the way they build software, we mean it — because it's how we build our own. Here's an honest look at the stack, the workflow and the guardrails behind howtech.uk, and what it means for the work we do for clients.
The stack at a glance
Four layers, each chosen so AI can do what it's good at without weakening what matters.
Why we built it this way
howtech.uk is a React single-page app, but it has the SEO behaviour of a static site and the operational profile of a small platform. We split the work along the seams that AI handles best: Lovable.dev owns the visible component layer — hero sections, service pages, forms, animations — and Claude Code owns the infrastructure: the Cloudflare Worker, the prerender pipeline, Supabase edge functions, GitHub Actions and the route registry that ties everything together.
That separation matters. Component code changes constantly as we iterate on copy and design, so we want the fastest possible feedback loop there. Infrastructure code changes rarely but has to be exactly right — so we use a different AI workflow with deeper context, code review and test coverage. Same idea we recommend to clients: pick the right tool for the right surface area, and don't pretend one workflow fits everything.
The site runs across three domains — howtech.uk, howell.uk and howell.co.uk — with a single canonical host and a Cloudflare Worker handling redirects at the edge. AI helped us design, test and ship that worker in a fraction of the time it would have taken otherwise.
What AI is genuinely good at here
Four places where AI has already paid for itself many times over on this build.
Iterating on copy and components
Lovable lets us turn a sketch or a paragraph of intent into a working, themed, accessible component in minutes. The same loop that used to take a developer half a day is now a short conversation.
Single source of truth for SEO
Every route, title, description and JSON-LD entry on the site is generated from one config file. AI is excellent at keeping ~60 routes consistent — and at updating them all at once when our messaging changes.
The boring-but-critical glue
Edge functions, prerender scripts, the Cloudflare Worker that routes bots to static snapshots — AI writes this kind of infrastructure code quickly and predictably, so we can focus engineering time where it matters.
Tests and accessibility at scale
We use AI to backfill unit, end-to-end and axe-core accessibility tests across dozens of pages. It's a force multiplier for the quality work that humans typically run out of time for.
What we don't trust AI to do alone
The flip side of moving fast with AI is being deliberate about where humans stay firmly in the loop. These are the guardrails that turn an AI-assisted build into a production-grade site, not a prototype with good intentions.
- •Every change is reviewed by a human before it goes live — AI is a collaborator, not an autopilot.
- •Security headers, CSP, consent gating and analytics rules are curated by hand and protected by tests.
- •Supabase access is locked down with row-level security and a separate roles table — never trusted to client code.
- •Accessibility is gated in CI with axe-core; visual and contrast issues are tracked, not silently ignored.
- •A documented project memory keeps the AI on-brand and on-pattern across sessions, so style and tone don't drift.
Built to be found by humans and AI
A React SPA on its own is a blank page to most crawlers. To fix that, every public route on howtech.uk is prerendered to static HTML during deployment, uploaded to Cloudflare R2, and served from the edge. A small Cloudflare Worker inspects each incoming request: if it's a known bot, an AI crawler, a text-based browser or a no-JavaScript client, we serve the static snapshot. Real browsers get the full SPA.
That's the part most consultancies aren't doing yet, and it's why our pages show up cleanly in ChatGPT, Perplexity, Google AI Overviews and traditional search alike. We also publish an llms.txt manifest so AI systems can discover the site's structure directly — generated automatically from the same route registry that produces our sitemap.
GEO — generative engine optimisation — isn't magic. It's the same discipline as good SEO, applied to a new generation of readers. AI helps us implement it consistently across every page on the site.
What this means for your business
The same approach scales beyond a marketing site. We use this stack — and the same blend of AI, infrastructure and human review — to build customer portals, internal tools and bespoke applications for clients. AI cuts delivery time dramatically; the wrapper of governance, testing, accessibility and security is what makes the result production-grade.
If you're considering AI-assisted development for your own business, the questions worth asking aren't "should we use AI?" but "where in our delivery pipeline does AI add the most value, and what guardrails do we need around it?" We'd be happy to help you answer those in your own context.
Talk to us about AI-assisted delivery
Whether you're modernising a marketing site, building an internal tool, or planning a larger digital product, we can help you adopt AI-assisted development safely and effectively.
Related insights
From opportunity to production: a practical AI adoption guide for SMEs.
A pragmatic, phased roadmap for small businesses adopting AI.
How to safely and effectively integrate AI capabilities across Microsoft 365.