The Personal Agent war has begun just like that. It’s a new battle for the entry point, with the survival of major tech companies at stake. I see four moats:
-
Context: Hardware and long-term memory are barriers to entry. But context is also a compliance liability—the jurisdiction where data resides directly determines whether a company can be acquired or expand internationally.
-
Distribution: Major companies have system-level entry points. Startups only have a chance at independent entry points through hardware or specific vertical use cases.
-
Action: Executing tasks on users’ behalf requires access to external systems such as e-commerce, payments, and browsers. These are precisely the areas platform owners are most reluctant to open.
-
Cost: Large price cuts by model providers show that inference costs are being pushed down upstream, which is good for the application layer. But once subsidies shrink, products that depend on them will be the first to suffer.
Software agents will eventually converge. To break out of that sameness, companies will have to build hardware. Hardware can also capture more context—valuable data from the real world. So I’m very curious about what kinds of hardware each company will release.