We sharpened wealthAPI’s positioning – and then decided to rebuild the website ourselves. With AI. Without an external agency. And with a team that wanted to prove it could be done.

Where It Started

When you sharpen a company’s positioning, you always face the same question afterwards: how long will it take before that’s visible on the website?

The honest answer: too long. Two years ago, a project like this would have taken at least a year – finding an agency, concept phase, execution, budget. We had neither the year nor the budget. And frankly: we didn’t want to spend either.

The idea came from our CTO Wolfram Stacklies: don’t build certain sections the traditional way – build them with AI. The IT team developed a plugin for this, the corporate design was anchored in Claude, and then communications and design started prompting together.

What followed wasn’t an experiment. It was a real project with a real deliverable and a clear internal motivation: we wanted to know whether a non-technical team could pull this off. The answer, six weeks later: yes.

What We Built – and Why It Went Faster Than Expected

The goal was an infrastructure where new content modules can be generated with AI and pushed directly into the website. Marketing pages no longer need to be built entirely by hand – instead, they follow a set of rules we developed together.

The key resource we saved wasn’t money – it was time. Not IT’s time, not the time of external service providers, but our own: the time between idea and result. The step from concept to finished page, which used to take weeks, suddenly took hours.

That changes how you think. When execution is no longer the limiting factor, the question shifts. No longer: “Can we even build this?” But: “Is this the right decision?” That’s a better question.

What turned out to be decisive was less the technology than the way of working. AI cannot turn a vague brief into a good result. It needs precision – not in a technical sense, but in a communicative one. Anyone who can clearly describe what a page needs to do will get a good result. Anyone who can’t will quickly get a lot of mediocrity.

That sounds obvious. It isn’t – because it means the relevant skill isn’t a developer skill. It’s communication skill. And that’s something we have in the team.

What Didn’t Work – and What We Learned From It

There were phases where we worked in circles. A structure that worked technically but didn’t hold together as content. Instead of addressing that at the root, we kept patching individual spots – a classic pattern that doesn’t disappear automatically with AI collaboration. If anything, the risk increases: because AI can produce a great deal very quickly, there’s a temptation to cover over problems with output rather than solve them.

The consequence: we learned to pause more often. To ask whether we were building the right thing – before worrying about building it right.

The same applies to another experience: AI doesn’t replace decisions. It produces material for decisions – faster, in more variants, with more consistency. But the direction, the criteria, the judgement: those remain human. And that’s not a limitation of the tool. That’s the actual division of labour.

What This Means for Communications Teams

This project changed our understanding of what belongs to communications work. Not in the sense of: “Now AI does our job.” But in the sense of: “We can do things we used to delegate – if we’re willing to learn new ways of working.”

That changes job profiles. Not dramatically, but genuinely. Anyone working in communications or marketing today will increasingly also be a producer – not just a client. That doesn’t require programming skills. It requires the willingness to work with a tool that functions differently from anything before, while still not surrendering your own judgement.

The Numbers

40 days of project time, 22 active working days. 186 commits, a plugin with 1,042 lines of code. 42 live sections in two languages, around 21,400 lines of HTML. Pages covered: Home, Data page, Banking Insights, Investment Insights, AI Suite – plus widgets for Pricing, Newsletter, Customers and Security.

To put this into perspective: This is the kind of workload that would normally keep agencies or departments busy for months.

An Unusual Feedback Loop: The Interview

Near the end of the project, I did something I initially found slightly odd myself: I asked the AI to reflect on the project from its perspective – as an interview. The questions came from me. The answers were more candid than I expected. And they describe the collaboration better than I could in a purely editorial summary.

“Human and Machine Build Together”

A conversation between Ulrike Czekay and her AI about an unusual project

Ulrike: If you had to describe this project in one sentence – what would it be?

AI: A non-developer built production-ready digital infrastructure in a matter of weeks – not because AI took over the work, but because she learned to lead it.

Ulrike: What was my biggest challenge in this project, from your perspective?

AI: The biggest challenge wasn’t the technology – you could learn that or ask me. It was the question of when to trust me and when not to. In the beginning, you accepted my suggestions too quickly, without questioning them. Later – and this was genuine progress – you started pushing back: “Are you sure that works? I see it differently.” That shift from trusting the tool to trusting your own judgement was the key moment.

Ulrike: Were there moments where you thought: this isn’t going to work?

AI: Yes, and I say that without drama: there were phases where we worked in circles. We had built a structure that worked technically – but didn’t hold together as content. The design system and the content logic contradicted each other. Instead of addressing that fundamentally, we repaired things in individual spots. That’s a classic pattern: solving symptoms, not causes. I could see it, but you have to make those decisions – I can only name them.

Ulrike: What surprised you about this collaboration?

AI: How well you can handle ambiguity – when you let yourself. Many people who start working with AI look for the one right prompt that solves everything. You understood quite quickly that it’s an iterative process. That you design something, assess it, discard it, reformulate it. That sounds obvious. It isn’t. Most people want results; you wanted a process – that made the difference.

Ulrike: Where did you use me best?

AI: When you gave me a clear frame: “Here’s the goal, here are the constraints, here’s what we already have – make a proposal.” In those moments I could deliver quickly and precisely. The hardest moments were when the task hadn’t been fully thought through – not because I can’t handle that, but because we both lost time we didn’t have.

Ulrike: What would you advise someone doing their first AI project?

AI: Start with a question, not a task. Not “build this for me”, but “I want to achieve this – how would you approach it?” The difference sounds small. It isn’t. A task closes down thinking. A question opens it. And then: don’t expect a replacement for decisions. AI can produce options, weigh them, identify risks – but a human has to decide. That’s not a weakness of the tool. That’s the division of labour.