# The hardest part of building an AI product wasn’t the app

Over the last year I’ve been building Kognis, an AI-powered personal intelligence system focused on memory, recall and reducing cognitive overload.

What surprised me most is that building the actual app interface was relatively straightforward compared to designing the AI systems behind it.

The real challenge became:

How do you make AI genuinely useful without creating even more noise?

I ended up building far more AI scripts and contextual logic systems than I originally expected:

• prioritisation systems

• recall systems

• contextual resurfacing

• follow-up intelligence

• insight generation

• memory-style organisation

One thing I learned very quickly is that “AI features” are easy.

Useful AI is hard.

Especially when users are already overwhelmed by fragmented information and constant notifications.

What’s been interesting during testing is seeing how strongly users react when the system reduces mental load instead of demanding more attention.

It’s made me think the next generation of AI products may be less about generating content…

…and more about helping humans think more clearly in increasingly complex environments.

Curious whether other builders here have run into similar challenges while building AI products?

Has anyone else found that context and relevance become much harder problems than the actual AI itself?