Your customer decides to stop using your customer relationship management (CRM) software at 2 a.m. Why? The sales rep realized that your $30,000 annual contract for your CRM only gives you $12,000 in value. There was no meeting, no negotiation. Because the sales rep was an agent.
Agents are not loyal. They are rational actors. They look at the numbers in milliseconds and make changes whenever it makes sense for the business, even if it is the middle of the night. They are, in a way, ruthless.
Your software can be easy to use and look good, but AI agents neither see nor care. Some companies will become winners by building headless architecture, which is software built for machine-to-machine communications. These companies won't have any human users, only agents.
Other businesses that will do well are in areas where you can't move fast and make mistakes, like regulatory approval, banking, and compliance systems. These companies aren't competing to be the most sophisticated, but to be the most efficient at things like having wire transfers delivered securely.
Over the next few years, the companies that own this layer will have an edge in gathering the intelligence about who should do what, and when. As the capabilities of the models converge and start to make the same sorts of decisions, companies will compete less on having the best model and more on the systems that connect those decisions to real-world outcomes. Some of these essential systems manage task routing, data access, workflow orchestration, and rule enforcement. Companies that are serving specific customers with specific use cases like these will win over broad agentic use cases.
These business models are like toll roads. If you don't want to pay, you've got to build your own bridge—and that could take years and cost hundreds of millions for compliance. That's why in the end, the less glamorous work may be the most rewarding.
