Claude’s recent back to back updates made something very clear to me: Software is no longer meant only for humans.

For years, SaaS products were designed around a predictable assumption.

A human logs in, clicks around, and uses the product a few times a day. Pricing models were built around that behavior too.

But with tools like Claude writing code, executing workflows and calling APIs autonomously, that assumption is breaking.

Software is now being invoked by agents. An agent might run a workflow ten times in a minute. Or once in a week. It might spin up five tools for a single task and never touch them again.

The usage pattern becomes bursty, unpredictable, and completely detached from the idea of “seats”. This changes how infrastructure needs to be built.

Rate limits, entitlement checks, credit systems and billing logic now need to operate in real time. The system has to understand not just who the customer is, but what the agent is doing and how resources are being consumed.

This is why traditional subscription models are starting to feel increasingly misaligned for AI-native products.

When agents become the primary users of software, access-based pricing stops making sense. What matters instead is execution, consumption, and outcomes.

The companies that recognize this shift early will design systems that are built for agents from day one.

And the rest will spend the next few years slowly rewriting their billing infrastructure.