The keyboard exists because typewriters jammed.

The command line exists because computers couldn't process much else.

The mouse, menus, forms, touchscreens, voice assistants, and even prompts were all brilliant responses to the constraints of their era.

But human interfaces somehow outlive the constraints that created them, but AI should help us escape them.

In this talk, I'll briefly trace the evolution of human-computer interaction from loom punch cards and command lines to modern AI systems through a different lens: not what each interface enabled, but the constraints that shaped it. We'll connect relatable examples such as IDE tooling to address coding complexity; why humans need continuously rebuild missing context through emojis, reactions, vs multimodal communication; why "natural language" interfaces still feel surprisingly unnatural; and why even the best AI systems still struggle with things humans handle effortlessly—interruptions, attention, turn-taking, shared context, and participation in real conversations.

The history of computing can be viewed as a series of increasingly expressive ways for humans to represent intent to machines. But representation is not participation, and certainly not rich and expressive.

Common wisdom suggests AI is an intelligence technology. I increasingly think of it as an interface technology.

For seventy years we've improved how humans encode intent for computers. AI may be the first technology that can meaningfully improve how computers understand humans. This shift changes more than software. It changes the interface between humans and computers.

The question isn't whether AI gets smarter, it is which constraints we stop accepting as fundamental.