Weak AppIntent Incentives Limit Assistants
Sam Hall, CEO of Wafer, on AI agent form factors
The weak incentive to expose AppIntents is what keeps assistant based AI from becoming a true app replacement. Developers only wire up a narrow set of actions when assistant traffic is small and hard to monetize, so assistants can order an Uber or start a reservation, but they still cannot reliably cover the long tail of real user workflows. That limitation is exactly why some companies go lower into the OS, or into dedicated hardware, to watch and learn actions directly.
-
AppIntents and similar assistant hooks are explicit, developer maintained handoff points. They are safer and easier for platforms to govern, but they only expose what an app decides to publish, which makes assistants broad in theory and shallow in practice.
-
Rabbit and related hardware tried the opposite path, skipping formal APIs and interacting with software more like a human user. The upside is broader coverage and novel interfaces. The downside is fragility, because every broken login, UI change, or edge case can cause the workflow to fail.
-
Dedicated AI hardware makes the most sense when the device captures context a phone does not naturally capture, like always on audio, camera, or glanceable interaction. For general purpose action taking, the phone still has the better starting position because it already has identity, apps, payments, and daily usage.
The market is moving toward a split model. Phones and OS level software will absorb most general assistant behavior, while AI specific hardware survives where form factor creates a new data advantage, like ambient memory or hands free capture. The winning products will be the ones that combine deep context with reliable action coverage, not just a new device shell.