AI

Everyone in this industry now says something about AI. The difference is not whether you use it, but whether you can say where it lets you down. Two questions come back here most often: what vibecoding actually delivers, and how you get found in AI answers.

For us, AI is a standard part of the development workflow. Not to generate software without developers, but to let experienced developers work faster: writing code, analysing existing code, making tests, preparing refactoring. We work with models including Claude, OpenAI and Grok, often through Cursor, and we keep that deliberately model-independent — the market moves too fast to tie ourselves to one supplier. Over the coming years we want to experiment more with self-hosted models, for more control over data, cost and performance.

What AI changes, and what it makes more important

Custom work was the expensive option for a long time, partly because so much development time went into repetitive work. Now that part of that cost falls away, the line between buy and build shifts. Especially for the smaller pieces of software that used to sit between a spreadsheet and a large project.

What does not change: architecture, product choices, integrations, and understanding how a business works. As code gets cheaper to write, the choice of what you build weighs more. Building the wrong thing faster is not a win.

Responsibility for what we deliver stays with our team. QA tests what gets built, whether a human or a model wrote the first version.

The second AI question: will I still get found?

We get that more and more, and fairly: more buyers now start with an assistant instead of Google.

Our position in one paragraph. The mechanics that work are largely the same as before, and they are unglamorous: clear factual content that answers a real question, structured data, the same company information everywhere, pages that are fast enough. What has changed: vague content no longer works. And the part nobody knows is how these systems weight sources — anyone selling you a guaranteed method in 2026 is selling you a guess.

What you will find here

• Getting picked up in AI answers: what works, what nobody knows, and what we measure ourselves
• Vibe coding versus software that has to last five years
• What happens to your data in third-party models

Questions we get a lot

What does AI cost in a project? AI usage runs through our infrastructure, so we pass on the capacity used at cost plus 20%. It is the only line we take a margin on; every other piece of software and hosting you buy directly. Pricing

Is structured data still needed if AI reads everything anyway? Yes, and more than before. It is the boring part that still works: it makes explicit what a machine would otherwise have to guess.

What happens to our code and data in those models? A client should get clarity on that, not trust. What we do and do not put in a model is in the article above.

Do you build AI into what we have you make? If it genuinely adds something. At scale, token usage is an architecture decision, not a detail. Platforms

Where this continues on the site

The build describes how AI sits in our sprints. Growth covers findability as a service, including the AI side.