Platform Incentives Limit AI Agents

This is a platform incentive problem, not a model capability problem. Google can technically push Android further toward an agent that completes tasks across apps, but Play still makes money when users open an app, buy a subscription, or pay for digital goods through Google billing. If an assistant starts handling the transaction in the background, the store risks becoming less central to discovery, checkout, and monetization.

• Google says Play funds itself through service fees on paid apps and digital purchases, and that most paying developers are on fees of 15% or less. That means Play is not just a catalog, it is a tollbooth tied directly to transaction flow. Anything that moves user intent outside that flow pressures the economics.
  2 google 3 google
• The practical bottleneck is app permissions. In the Wafer interview, Sam Hall describes current assistants as limited to actions developers explicitly expose through AppIntents. That keeps Google and Apple in a safer posture, because the assistant can request narrow functions like book a ride, instead of freely reading app data and routing around the app UI.
  1 sacra
• Android is also more valuable to Google as ecosystem leverage than as handset share. Pixel held about 3% of the US smartphone market in 2025, while Samsung had 23% and Apple 56% in the same source. That supports the broader point that Google uses Android to control distribution and defaults across a much larger OEM base than its own hardware footprint.
  4 counterpointresearch 5 canalys

The next step is likely not a sudden move to a phone without apps, but a gradual shift where assistants handle more high intent actions while apps become back ends exposing structured functions. The winners will be the platforms and startups that can capture user intent at the OS layer without breaking the billing and developer relationships that still fund mobile software.