Every few years the industry finds a way to lower the barrier to building software. A few years ago it was low-code platforms like Power Apps, Power Automate and Zapier, which let operations managers, marketing managers and finance leads build the tools they needed without waiting on a developer.

Now AI is making the same pitch, only with an even lower barrier. Describe what you want in plain English, and you can be surprisingly close to a working tool in minutes.

But what isn’t in the sales pitch is what happens two years later.

What actually happened with low-code

Low-code solved a real problem. Developers were expensive, business teams had more demand than IT could support, so business teams built a lot.

Fast forward to today, and most mid-market organisations are staring at a long tail of tools and applications that are poorly managed, poorly monitored, sometimes built by people who have already left the business, and definitely still running. Nobody is quite sure what will break if they turn them off. The licensing and hosting costs keep adding up, quietly, in the background.

The lesson is that making something easy to build didn’t make it easy to own. That’s the lesson we’re about to repeat with AI, only at a bigger scale.

Why AI is a bigger version of the same problem

Low-code lowered the barrier to whoever was willing to learn a platform. AI is lowering it further. Your team doesn’t even need to learn the platform, they just need to describe what they want.

Which means a lot more people can build things a lot faster, with a lot less visibility.

If you’re wondering whether your existing investment in Power Apps or Power Automate is now wasted, it isn’t. If the tool is still hitting the mark, still being used, and was implemented with proper governance, there’s no reason to rebuild it with AI. AI isn’t actually replacing these platforms either. Microsoft is embedding AI directly into Power Apps. So low-code and AI are converging.

The pattern we already saw with data

We’ve been here before, with dashboards. Giving people the ability to create dashboards more easily didn’t remove the problem of trust and confidence in the data underneath. If anything, it made it worse.

Apps are about to go through the same cycle. Anyone can now build one. Not everyone can take one to production and maintain it.

Three things to get in place before it gets out of hand

1. Get ahead of who’s building. With AI sitting inside ChatGPT, Claude and Copilot, potential builders now include anyone with a login. Before that list gets longer, document what good looks like: where apps live, what they point to, how they should look, applied to anyone building anything.
2. Every app needs a clear owner, a reason for existing, and a time it can be switched off. If nobody can name who owns it and why, it probably shouldn’t be running.
3. Have a path to productionise the ones that matter. When an app becomes widely used or business-critical, hand it over to the technical team to be properly productionised. Business users build for themselves. Once others rely on it, it isn’t a business user’s app anymore.

The point

Low-code didn’t create a technology problem. It created an accountability problem. AI is going to do the same thing, at a bigger scale.

The organisations that come out of this looking healthy won’t be the ones with the most apps. They’ll be the ones with governance in place before the building started.