From a design to a working website, with AI.
My own website is one of the systems I build with AI. The useful part is the connection between the design, the code and what a visitor can actually do. Here is how that connection works in the byImprint build.
A design is the beginning of the work.
byImprint’s design system lives in Paper. It sets the typography, spacing, colour, navigation and page families. I use AI assistance to work through those designs and translate them into the Next.js website, keeping the reusable parts together.
The work is easier to inspect when a service page, a guide and a case study share a deliberate system. A change to the navigation should reach each of them. A phone layout should have been considered before it becomes a surprise in the browser.
Four things I ask the workflow to produce.
For the September 2026 inbound build, the brief covered service pages, practical guides and a workflow-mapping tool. The workflow needed to produce more than attractive screens.
- A usable design
- Desktop and phone layouts in Paper, with the real copy and visible states.
- Shared implementation
- Reusable React components and structured content, so common changes have a clear home.
- Working behaviour
- Links, navigation, worksheet controls, download and print behaviour checked in the browser.
- A reviewable result
- A local production build that can be inspected before any public release.
The checks catch different kinds of mistakes.
I inspect the rendered pages at desktop and phone widths. I follow the links from the home page and check that useful pages can be found. For the workflow tool, I enter an example, export it and check the print result. A page that looks right can still lose the user’s work or lead to a dead end.
The earlier inbound build passed 58 focused checks and a 50-viewport review. Those numbers describe that local build, not the quality of every future change. New behaviour gets its own check; a passing test does not replace looking at the page.
Where my judgment comes in.
I choose what the page needs to say, what evidence supports it and whether the result feels right. I review the design and the actual implementation together. If a section is weak, I change the section; adding more pages will not fix it.
AI helps with implementation and iteration. It can also repeat a mistake across an entire family of pages. Shared components reduce that risk, while visual review catches what code checks cannot. Both are necessary.
What this demonstrates.
This is byImprint’s own website workflow, reviewed in a local build on 9 September 2026. It shows how I connect AI-assisted design and development to an inspectable result. The latest expansion still requires a separate public release.
It is not a client savings claim or a timed comparison with unaided development. For your business, we would agree the scope, the acceptance checks and the result worth measuring before we build.
