With AI you build it yourself, but not without know-how

With AI you can build a dashboard, a customer portal or a handy tool for your team in a short time. You describe what you want and the AI makes it. For many business owners that feels like a way to skip the bill from a software company.

But a tool that works on your own laptop is not yet a tool your team can rely on every day. It has to run somewhere and be reachable for the right people. And it has to stay secure, even when something changes a year from now in the software it runs on.

On 21 October two serious leaks in PHP became known. PHP is the programming language that many websites run on, from web shops to WordPress sites.

According to the Canadian Centre for Cyber Security, these leaks let someone run software on a website's server without logging in. The fix is an update of PHP itself.

According to the Dutch government, the NCSC advised owners of PHP websites to weigh the risk themselves and take their site offline if needed until the update was there. That is why a number of government websites went offline as a precaution. At that moment there were no signs of abuse yet.

At the same time it is often said that AI makes building software so cheap that you no longer need a software company. For the building itself that is more and more often true.

DORA, the research team at Google Cloud, looked this year at what AI delivers when building software. According to DORA, AI works as an amplifier. What is already well organised gets better. What is messy gets messier too.

So according to the report, the biggest return does not come from the tools themselves. It comes from what surrounds them: a well set-up place to build on, clear agreements about the work and a team that is aligned. Without that foundation the gain stays stuck in separate pieces of work. The report is about teams that build software, not about a business owner who makes a tool themselves.

It is true that you can build a lot yourself with AI. AI is good at making a tool, a dashboard or a small platform. For that you no longer always have to hire a software company.

But building is only part of the work. A tool has to run somewhere, with an address and access for the right people. And that place has to be updated when there is a leak, as there was this week with PHP. AI does not watch for that by itself. Someone has to know where the tool runs and what needs updating.

Starting from scratch is also the easiest thing for AI. Building further on what is already there, such as your own systems and earlier tools, takes clear instructions and the right model. Without that knowledge AI sometimes picks an approach that does not fit what is already there.

What that looks like in practice is shown by the platform migration for Ad Your Service. The platform's website and background tasks ran at different hosting providers. They were moved to one place in steps, so the team only has to know one way of working. Every change also gets a test environment of its own, so a mistake shows up before clients are bothered by it.

This approach does ask something of you. You have to make time to write down what you build and to keep up with updates, or ask someone to do that. For a tool that only you use now and then, that is often more than needed. For a tool your team works with every day, it is part of the job.