Skip to content
Menu

How we build

The engineering practices behind our work, how we use AI coding agents, and how we keep AI systems reliable and responsible.

AI changes how software is made and what software can do. It does not change who is responsible for the result. This page describes the practices we follow on every project, how AI coding agents fit into them, and what we do to make AI systems reliable.

Engineering practices

These apply to every project, with or without AI.

  • Requirements in writing. Before any code, we agree what the software must do, for whom, and how we will know it works.
  • Architecture with reasons. We choose a structure that fits your scale, team and budget, and write down why, so later decisions have context.
  • Automated tests. Important behaviour is covered by automated tests. What we test, and how thoroughly, depends on the project and its risks.
  • Code review. Changes are reviewed before they become part of the product, with more care where more is at stake.
  • Careful releases. Changes go live in small steps, with a way back if something goes wrong.
  • Security from the start. Access control, protected keys and up-to-date dependencies are part of the work, not a final check.
  • Documentation and handover. Setup, decisions and operation are written down, so that we, your team or another team can maintain the system.

Development with AI coding agents

We use AI coding agents in our development work, and we have built complete applications this way, including our own product, Cookamigo. The agents work within the practices above, not instead of them.

  1. We decide. We define the requirements and agree on the architecture. Agents can propose options. We compare them and choose.
  2. Agents implement. The code is written in small steps that follow the agreed design and the conventions of the project.
  3. The same checks apply. Code written by agents goes through the same tests and review as code written by hand.
  4. We verify. We run the software as a user would, in the browser or on the device, and correct what we find.

What this does not mean. AI does not write perfect code, and the work is not autonomous. Agents make mistakes that look plausible, which is why tests and review matter. We do not promise that a project will be many times faster. What changes is where the engineering time goes: more into requirements, design and checking, less into writing routine code.

Making AI systems reliable

No AI system is right every time. We use testing, monitoring and feedback from users to find problems early and to improve the system over time.

  • Grounded answers. The system answers from your documents and data and shows where each answer comes from, so people can check it.
  • Missing information. The system is built to say when the sources do not contain the answer, instead of guessing, and to pass the question to a person where that makes sense.
  • Evaluation and testing. Before launch we agree on a set of real test cases and measure the system against them. The same tests run again after every change to prompts, models or data.
  • Monitoring and improvement. We set up monitoring of quality, errors and cost, and a simple way for users to report wrong answers, so the system can be improved based on real use.

Responsible use of AI

  • Human oversight. Where a mistake would be costly, AI prepares the work and a person decides.
  • Data privacy. We agree which data the system may use, where it is processed and who can see it. The access rules of your existing systems still apply.
  • Security. We treat everything the model reads as untrusted input, limit what the system is allowed to do and protect the keys and data it uses.
  • Clear limits. We make it clear to users that they are working with AI and what the system is not meant to do.

Have a problem AI might solve?

Describe it in a few lines. The first conversation is free and non-binding, and we usually reply within one to two business days.

Let's talk