AI Can Build Apps—But Can It Understand Your Business?

With the news about GitHub Spark and talk of Power Apps generative pages, the paradigm of how AI might be involved in app creation seems to be shifting again to create a clearer path from natural language input to app output.

I am not a coder, but in limited testing of some of these types of capabilities so far, it seems to me that AI-based app building suffers from the same basic weakness that AI-based knowledge worker support does too: your success is entirely limited by context.

The reason why tools like Microsoft 365 Copilot or ChatGPT's SharePoint connected Deep Research are powerful is because they plug right into contextual information that makes their outputs more relevant. If you just ask ChatGPT - or even Microsoft 365 Copilot Chat - specific questions about your business without adding necessary specific context, the results will be bland and disconnected from the reality of how you work.

In general, custom business applications are driven by context too. If you have a process in your business that you do much the same as many other businesses, then you don't need a custom app, there will be many off the shelf solutions that allow you to complete that vanilla process successfully and integrate with a bunch of other apps too. It's only where you do things uniquely or significantly differently to the norm that a custom app makes sense over a prebuilt one.

The problem with natural language input to app output is conveying this context. I've found that fairly mundane requirements you can find anywhere are easy for tools like the AI app builder in Power Apps to work out, but when you step outside of that standard, general, vanilla, way of doing things, it becomes far more complicated.

First, truly understanding what the real requirements are is a specialist task in itself. Most knowledge workers have a fairly limited ability to concisely convey an appropriate set of functional requirements without prompting and workshopping.

Second, even when those requirements are clear, the complexity for the AI platforms seems to come in the non-standard needs. You end up with data models that either make no sense or simply are not structured to deliver the requirement that has been shared. The logical leaps that consultants and developers make to turn ambiguity into success are not yet that well integrated into many of the AI solutions.

These are truly interesting technologies but, in my opinion, there is a long road ahead before AI is able to exceed the capabilities of skilled custom app builders. There are lots of steps to success here, and the actual building part is a comparatively small aspect.

❓ I'd be interested to learn what those who are more hands on with app building every day think of this issue. Am I overemphasizing the importance of these human interventions? Or should businesses with unique needs still start with human experts rather than AI help?

Next
Next

AI Isn’t Free Anymore—Copilot’s New Limits Say So