You type: “Add the new collection in German as well, and build a landing page for the trade fair.” That is the entire brief. A few minutes later it is ready: translated across every field, added to the navigation and the sitemap, with a preview link — waiting for your approval.
No ticket with the agency. No queue. No clicking through a page builder. This works because the website is built differently underneath: WordPress as the content source, a Next.js front end in Git, and Claude as the layer allowed to operate both.
An ordinary website cannot be operated — by anyone, AI included
On a conventional WordPress site your content is locked inside visual blocks. A piece of text does not sit in a field as text; it sits as formatting, in a column, in a section, in a template. The best an AI can do is suggest a change that you then retype yourself. That is not operating a website. That is dictation.
Headless turns that around. Content becomes data behind an API — fields, languages, statuses. The front end becomes code in Git. Which means: anything a person can click in your website, an agent can call. And every change is a version you can roll back.
- Conventional: content as formatting — text is entangled with layout. It can only be changed by clicking, so only by people, one screen at a time.
- Headless: content as data — title, intro, specifications and SEO text per language, in separate fields. Readable and writable through the API, in bulk.
- In Git: change as version — every adjustment to the site is a commit with a preview. That is precisely why AI can be trusted with it: if it goes wrong, you are one command away from yesterday.
What you ask for, and what happens next
Five jobs that take a week on an ordinary site and arrive as an agency invoice.
- “Add these twelve new products in German too.”
Title, intro, body copy, SEO title and meta description translated per product, written into the German field set and marked for review. They only go live once someone approves them. - “Build a landing page for the Frankfurt trade fair with these six products.”
A new page in every language, with hreflang references, added to the sitemap, linked from the relevant category pages — as a draft on staging, with a link you can forward. - “Update the specifications; they are in this spreadsheet.”
Forty products updated through the API, datasheets regenerated, plus a list of differences so you can see exactly what changed before you sign off. - “Why is this page slipping in Google?”
An analysis of titles, headings, internal links and competing pages on your own site, with a concrete proposal. Implemented as soon as you approve — not delivered as a report that gathers dust. - “Add the new range to the menu and the filters.”
Navigation, filter values and overview pages updated, in a single branch, with a preview link. Don’t like it? Throw the branch away and nothing happened.
Three layers, two of which you already know
- WordPress stays. Your editors work in an environment your team already knows. The difference is in how it is set up: content lives in structured fields per language, not in formatted blocks. That is exactly what makes it machine-readable.
- The front end is code. A Next.js front end: statically rendered, extremely fast, multilingual, with SEO and hreflang properly handled. It lives in your own Git repository, with staging and production separated and a preview URL for every change.
- The AI layer may operate both. Claude is given access to the content API and to the repository — with the same bounded permissions you would give a new employee. Which content types, which fields, which environment: you decide that during setup.
Who presses publish? You do.
Everyone’s second question. The answer is in the setup, not in a promise.
- Proposals land on staging, never straight onto your live site.
- Every code change is a branch with a preview link and a review moment.
- Translations are marked for review; only approved content is published.
- Everything is in Git. Going back to yesterday is one command.
- Access is bounded by content type, by field, by environment — and revocable.
AI does the repetitive work here: translating, updating, rolling out, checking. The editorial judgement stays with a human, because that is not the part you want to outsource.
What it looks like in production: Atomis Lighting
Atomis Lighting, a manufacturer of architectural lighting, moved from a stalled WooCommerce site to this setup. Not as a pilot — this is their production site: headless WordPress with a purpose-built content engine, a Next.js front end, four languages, roughly 700 pages, 99 products and 36 projects, with automatically generated datasheets and a quote request that lands straight in the CMS.
Atomis is the example here, not the product. What you buy is the way of building — on your content, your brand, your domain.
When this pays off — and when it is a waste of money
This is for you if
- you have a lot of content that changes regularly — a catalogue, projects, knowledge articles;
- you work in several languages, or intend to;
- your marketing team is small and your ambition is not;
- you depend on an agency for every change;
- speed and findability carry commercial weight.
Do not do this if
- your site is ten pages and changes twice a year;
- your content is unstructured everywhere and is going to stay that way;
- nobody internally wants to own the website;
- you mainly want a new look — buy a theme instead.
From first conversation to a website you can operate
- Discovery — 2 weeks, €2,500, credited if we proceed.
We model your content: which types, which fields, which languages, which workflow. You get a technical plan and a fixed price for the build. If it goes no further, you still have a usable content model. - Build — 6 to 10 weeks, from €24,950.
Setting up the CMS, building the front end, migrating your existing content, separating staging and production, automated build validation. - The AI layer — €5,000, training included.
Setting up access, permissions and guardrails, documenting the content model so the AI understands it, and teaching your team what to ask for and what not to. This is the part that makes the rest worth having. - Support — €495 to €995 per month.
Hosting, monitoring, updates and further development. Plus a named contact for the moments when calling a human is simply easier.
Or as a single package
| Starter — one language, straightforward content structure, AI layer for copy and publishing | €15,000 – €20,000 |
| Professional — multiple languages, catalogue or project overview, migration, full AI workflow | €25,000 – €40,000 |
| Enterprise — integrations with PIM, ERP or CRM, complex workflows, multiple brands or domains | €40,000 – €75,000+ |
Frequently asked questions
Do I have to leave WordPress?
No. WordPress remains your editorial environment. Only the public side — what visitors see — is replaced by a faster, modern front end. Your editors keep a familiar screen.
Can the AI publish things on its own?
No. Proposals go to staging or are marked for review. Publishing is a human act and we keep it that way.
Which AI model does this use?
Claude, from Anthropic, because it currently handles code and structured content best. The setup is not welded to one model: your site is and remains WordPress plus a front end in Git.
What if I want to stop doing this in three years?
Then you keep everything. The content sits in your own WordPress, the code in your own Git repository. There is no platform in between to migrate away from — which is exactly why this setup was chosen.
Will this work on my existing website?
Usually, yes. The content migrates; the design is rebuilt. The messier your current content, the more work sits in the first step — that is what discovery is for.
How quickly can I go live?
Typically six to ten weeks after discovery, depending on the volume of content and the number of languages.
Does this put my content manager out of a job?
No — it takes away the retyping, the manual translating and the chasing of forgotten internal links. What remains is the work they were hired for: deciding what gets said.
Thirty minutes, and you see it happen yourself
No slides. We open a live site, I type an instruction, and you watch what comes out. After that you will know enough to judge whether this fits your organisation.
Related services
A website you can operate is one application of a broader shift: organising work so that AI can genuinely take it over. See also AI Process Implementation, digital transformation, ISO & Compliance Readiness and the Strategic Diagnostic Session.