One outcome for this is a process I’ve coined “Neo-institutionalization”.
Assuming the following is true:
- Long-horizon agent capabilities improve alongside reliability & observability.
- All-in marginal token cost trends towards zero.
- Agents can reliably produce economic outputs.
- Economic rents will migrate to the service with the soundest unit economics, who can deliver any good or service the most efficiently at the highest quality relative to AI spend.
In this scenario, value accrues to the a) provider of the intelligence & b) the institution that adopts the intelligence to deliver the good or service.
The application layer will be squeezed out. If agents can learn and improve recursively, they will need tools to be useful. Whoever can deliver the tools securely and at scale will be able to perform the cost efficient service at scale.
Those who create the tools for the agents to perform tasks cheaply & more reliably will win.
If interested, check out our AI Economic Viability Frontier: https://aevf-index.vercel.app/
Browse the task universe of the Economic Viability Frontier →