AI has made code cheap. With the latest AI tools, you can literally build any feature you want in no time (yes really). But this newfound superpower comes with a trap: feature bloat.

Just because you can build every feature doesn’t mean you should shove them all into your product.

Knowing what not to build is a critical skill - arguably more important than knowing what to build. It’s a skill that may ultimately determine the success or failure of your product, especially if you’re operating in a highly competitive market. Because reducing noise for your users increases focus. Focus increases engagement. And engagement… increases retention (and gets PMs bonuses).

The feature bloat epidemic

The feature bloat phenomenon is already evident in the majority of SaaS products. Suddenly everything is “AI-powered,” with AI features shoehorned into every corner of the product - even when it makes no sense. Take Snapchat, for instance. They pinned “My AI” to the top of user inboxes, forcing a chatbot into a space reserved for intimate conversations with friends. Users hated it. Or consider LinkedIn, now cluttered with AI-generated “collaborative articles” that often feel like low-effort engagement bait rather than professional insight. Or worst of all, Microsoft Windows 11, which has AI features scattered throughout the OS that most users don’t even know exist, let alone use.

Most of these AI features add zero value to the product itself. They exist because some PM wanted to appear “innovative” and “forward-thinking” to hit arbitrary KPIs. The result? Cluttered products that confuse users rather than help them.

Now, before you think I’m being negative about AI - absolutely not. Most blog posts on this site make it clear that I’m incredibly bullish on AI. I just don’t believe the world needs more bloated SaaS products with useless features shoved in everywhere without considering if they’re actually useful.

Learning from 37signals

The team at 37signals recently demonstrated this perfectly. They launched a new product called Fizzy and decided to open source it. From the git commit history, it’s clear that they initially added AI features to the product, realized it wasn’t adding any value, and decided to rip all of it out.

Perfect! It’s ok if AI is not useful in every single use case.

Everyone benefits in this case. The builders have fewer features to maintain and less complexity. The consumers get a more focused, valuable product without BS features forced in for some arbitrary purpose.

Unfortunately, this kind of restraint is rare in today’s AI craze.

Conclusion

In the age of AI, building is easier than ever. The hard part is restraint.

The best products aren’t the ones with the most features - they’re the ones that do fewer things exceptionally well. With that in mind, knowing what not to build might just be the most valuable skill you can develop as a product builder.