thank you for this, the connector ecosystem divergence is the most underrated story here. chatgpt going consumer super-app, claude going professional infrastructure. two very different bets
The most important insight here might be the one near at the end: the rankings increasingly undercount the AI products people actually use most.
A developer living in Claude Code and a knowledge worker dictating through Wispr are invisible in web traffic data. We're measuring AI adoption with a flashlight when we need a floodlight.
As AI shifts from "destination" to "feature," I wonder if the next edition of this list will need an entirely new methodology maybe something closer to compute-hours consumed or tasks completed rather than page visits and MAU (or even the dreaded # of tokens used!).
Great points. Over 90% of my use as Claude Code or in Xcode 26.3, maybe 8-9% via the Claude app on my iPad/iPhone, and Claude.ai on Safari once in a while. As AI becomes more of a feature on devices, determining usage via page visits isn't going to be of much use.
This was the biggest takeaway for me: AI is slowly stopping to feel like a separate product and starting to become the layer inside everything people already use. That shift may matter more than who is number one today, because once habit and workflow settle in, distribution can beat raw model quality.
Meta didn't just drop $2 Billion on Manus for fun. OpenAI didn't acquire OpenClaw by accident.
Big Tech knows the war for foundation models (AI that talks) is already commoditized. The trillion-dollar land grab happening right now is for the Execution Layer (AI that acts).
Whoever owns the agent that actually does the work, owns the user. We are watching the entire software stack get rewritten in real-time.
thank you for this, the connector ecosystem divergence is the most underrated story here. chatgpt going consumer super-app, claude going professional infrastructure. two very different bets
100%
The most important insight here might be the one near at the end: the rankings increasingly undercount the AI products people actually use most.
A developer living in Claude Code and a knowledge worker dictating through Wispr are invisible in web traffic data. We're measuring AI adoption with a flashlight when we need a floodlight.
As AI shifts from "destination" to "feature," I wonder if the next edition of this list will need an entirely new methodology maybe something closer to compute-hours consumed or tasks completed rather than page visits and MAU (or even the dreaded # of tokens used!).
Great points. Over 90% of my use as Claude Code or in Xcode 26.3, maybe 8-9% via the Claude app on my iPad/iPhone, and Claude.ai on Safari once in a while. As AI becomes more of a feature on devices, determining usage via page visits isn't going to be of much use.
I get the feeling that Apps and Web browsers may NOT have the future we once thought they did.
Thanks for curating this !!
Is the color wrong in terms of AI adoption in China?
Chinese users aren't avoiding those top AI products; it's just that they are likely using a VPN to access them LOL
Great article. Thanks for writing & sharing Olivia!
This was the biggest takeaway for me: AI is slowly stopping to feel like a separate product and starting to become the layer inside everything people already use. That shift may matter more than who is number one today, because once habit and workflow settle in, distribution can beat raw model quality.
A lot of these are juts not ai apps at all
I am so proud of leading the product for Spicychat and seeing the product in top 20. Let's goo
Meta didn't just drop $2 Billion on Manus for fun. OpenAI didn't acquire OpenClaw by accident.
Big Tech knows the war for foundation models (AI that talks) is already commoditized. The trillion-dollar land grab happening right now is for the Execution Layer (AI that acts).
Whoever owns the agent that actually does the work, owns the user. We are watching the entire software stack get rewritten in real-time.