Key Takeaways

  • Mike Krieger saw a strict power law at Instagram, where 80% of user activity remained on the main feed and all Explore tab improvements only shifted engagement from 10% to 15%.
  • Product teams build sidebars and new tabs because it lets them ship fast, but Ami Vora points out this practice forces users to figure out which tool fits their problem.
  • The gap between what frontier AI models can do and how people actually use them is a product design problem rather than a user education problem.
  • Strong product management in AI requires protecting the primary interface from clutter while integrating winning experiments directly into the core workflow.

The 80% Rule of Interface Real Estate

When product teams invent a new AI feature, their default move is to add a new tab, a sidebar drawer, or a dropdown menu. It feels safe. It avoids breaking existing habits while letting the team test something new.

Krieger warns that this design instinct clashes with how people actually use software. Looking back at his time building Instagram, user attention followed an unforgiving power law.

“The thing that we always saw at Instagram, there is a very strong power law in tab usage,” Krieger explained. “Main feed was 80% of the use. You can make Explore better and maybe you go from 10 to 15%, but main feed is where you land. That is your big thing.”

If a feature lives in a secondary tab, it fights for scraps of attention. Users do not explore complex navigation trees when they open an app to get work done. They sit on the main screen. If the core prompt box or primary workspace does not solve their problem, they leave. Splitting features across multiple entry points just dilutes focus.

The Trap of the Tool Picker

Giving users a menu of AI tools seems generous. In practice, it offloads product decisions onto the customer. When an interface asks someone to pick between a researcher agent, a code generator, and a writing assistant before they type a word, it creates friction.

Vora argues that users want an interface that handles the complexity behind the scenes.

“The user expects to walk up to something and hopefully not have to think about it too much,” Vora noted. “Right now it is easy to just transfer a bunch of cognitive load to the user and be like: here are some tools, why don't you choose?”

This dynamic explains why so many people get stuck using models for basic chat. Krieger calls this the gap between model ability and actual usage. “I often talk about the gap between what the models are capable of and what most folks are using them for. Not through any fault of their own. It is on us. We have got to build the products to make that possible.”

Bridging that gap does not mean adding twenty buttons to your navigation rail. It means building unified entry points where the system selects or blends the right tools without forcing the user to configure the backend.

Saying No to the Next Tab

Shipping experiments in isolated branches or temporary modules is fine for testing. The danger appears when those experiments stay isolated forever, turning your product into a collection of half-baked tools.

“Half of the job is saying no to another thing in the sidebar or another tab or another thing,” Krieger said. “And figuring out: can this be rolled in once we have proven it out, but not killing it so early?”

Great AI product design keeps the surface area tiny. Build the experiment, prove that it delivers real value, and then fold it back into the main workflow. If you cannot find a way to merge it into the primary interface without cluttering the screen, you probably should not ship it to production.

What to Do With This

Audit your product analytics this week to find your version of Instagram's 80% surface. Look at your secondary navigation items, sidebars, and sub-tabs: if any feature sits below 5% weekly active usage, do not redesign its tab. Either bring that capability directly into your main user input flow, or delete it entirely to reduce user decision fatigue.