AI stopped being a product. It's the base layer now
a16z released the third edition of their consumer AI report.
Three years ago everything was simple: there were AI products, and there was everything else. That boundary no longer exists. CapCut: 736 million MAU, with the most popular features AI-based. In Notion, usage of paid AI features grew from 20% to 50% in a year.
AI stopped being a product. It became the base layer inside any product. Below are the main trends I took away from the report:
1/ The race for "default AI" has only just started
ChatGPT = 900 million WAU. 8x more paid subscribers than Claude, 4x more than Gemini. The dominance is obvious. But for the first time the competitors are growing faster: Claude +200% YoY, Gemini +258%. And every fifth ChatGPT user already runs Gemini in parallel. The lead is definitely there. The loyalty and the habit are not.
Why it matters:
- The platforms are diverging in strategy. ChatGPT is building a super-app: shopping, bookings, ads, "Sign in with ChatGPT".
- Claude is going after power users: dev infrastructure, financial terminals, MCP connectors.
And it's now obvious this isn't the search wars, where one player takes 90%. It's more like iOS vs Android: two ecosystems, and both can survive.
- The key lock-in is context. The more the AI knows about you, the better it works, and the harder it is to leave. Whoever becomes the operating environment first, instead of a chat window, wins.
2/ One product = one AI task no longer works
The platforms are expanding and swallowing everything that used to be a separate product. In 2023, 7 of the 9 creator products in the a16z ranking were image generators. Now it's 3. Midjourney fell from the top 10 to 46th place. ChatGPT and Gemini built in their own image models, and for most users that's enough.
Why it matters:
- If your product does exactly one thing, sooner or later a platform will most likely add it as a feature. Two types survive: the super-app and the deep vertical with unique data. Everything in between will probably die. - It's going to go this way with roughly everything. Image generators are already dead, video is next. Music and voice are holding up better: Suno, ElevenLabs. The platforms haven't reached them yet, but they definitely will.
3/ From generating content to generating actions
The platforms swallowed image generation, text, and soon video. But content is passive output. The next step is AI that doesn't generate but acts: books, sends, fills in, coordinates.
OpenClaw, an open-source agent by a single developer from Austria, collected 68 thousand GitHub stars in a matter of weeks, overtook React and Linux, and got bought by OpenAI. Meta bought the agent platform Manus for ~$2 billion. Big companies don't build agents. They buy the ones already showing traction.
Why it matters:
- Agents are nowhere near user friendly yet. Adoption hasn't happened. But once users connect AI to their calendar, email and CRM, the switching cost shoots up. Connectors are the new lock-in.