AIWA "The One Thing" #07: AI Agents are real. They aren't evenly distributed. 50% live in one domain.

In June 2025, Gartner reported that 99% of technology vendors claiming to be Agentic, weren’t.

AI agents are real but they aren’t evenly distributed. Yet.

While it’s true that AI agents today are concentrated in software engineering, it’s also true that this concentration has generated results that have convinced CEOs and their boards where things are headed next.

Scaling AI agents beyond software engineering carries “higher stakes than fixing a bug.”

Three things enterprise leaders should take away.

1: Just because an AI Agent can do something doesn’t mean a human will enable it
Even the most advanced users aren’t necessarily using the technology to its complete potential. Anthropic found models capable of 5-hour tasks are operating for 45 minutes in practice. A reminder that testing “what agents are capable of in controlled settings” is good, but understanding “how people interact with agents in practice” is great.

2: Embrace + learn from your power users, investing in the infrastructure to do so
Power users can pull the future forward by leading from the front. Showing the rest of the organization what’s possible while demonstrating the gap between the hypothetical and what happens in real-world applications.

For example, in Anthropic’s research they found that Claude Code power users enable more autonomous work as they “gain experience with agents” but they “also interrupt more.”

Not less oversight, different oversight. That’s what scaling will look like.

3: Claude Cowork is going to accelerate domain expansion
Claude Cowork is to non-technical knowledge workers what Claude Code is to engineers. Democratizing access to the horsepower with enough abstraction to enable use across organizations.

One more thing.

This visual has been circulating across AI Twitter and once you see it, you can’t unsee it.

It’s gotta be at least directionally right, right? ;)