We're Using AI Wrong. The Chat Window Sucks

This is true for many AI use cases, but I don't agree that a chat interface is the same as exposing backend actions to a website user.

When working with an agent, you often need to see what it's doing so you can redirect or help it. Modern agents actually seem to hide more of this now. Codex, for example, often shows little more than "Run a command."

So the problem is real, but I don't think "We're Using AI Wrong" is the right framing. In cases where only the result matters, I agree. But there are many cases where seeing what the agent is doing is part of the interface, not an implementation detail.