May 16, 2024
By the middle of 2024, the prompt bar has become the most fashionable element in software. Type a request in plain language; the system interprets it and acts. It appears in operating systems, in browsers, in productivity tools, in the command surfaces of applications that previously hid their functions behind menus. The framing is usually that something genuinely new has arrived — an interface that finally lets you simply ask.
The framing is half right. The prompt bar is a meaningful development, but it is not new in the way its promoters suggest. Its direct ancestor was shipped in 1987, in a computer that sold for about six months and then disappeared. Jef Raskin designed it, and he called the mechanism LEAP. Reading the AI prompt bar as a descendant of LEAP clarifies both what it inherits and what it has given up.
LEAP, Briefly
LEAP was the navigation and command mechanism of the Canon Cat, the text-centered computer Raskin designed after leaving Apple. The Cat had no menus, no application launcher, no file browser. To do anything, you held a key and typed. The system found the nearest matching text and moved you to it instantly, character by character, with no results list to choose from. The same gesture found content, executed commands, and navigated, and it worked the same way on everything in the machine.
The principle behind LEAP was that the user should express intent in the most direct available form — characters of the thing they want — and that one uniform mechanism should serve every kind of finding and acting. Raskin spent the rest of his career arguing that this was the correct foundation for an interface and that the spatial, menu-driven graphical interface that won the market instead was a detour. The full argument is preserved in the Canon Cat documentation and in The Humane Interface.
What the Prompt Bar Inherits
The AI prompt bar carries forward LEAP’s central commitments more faithfully than any mainstream interface element before it.
It is uniform. A well-built prompt bar accepts a request to find a document, a request to perform a calculation, a request to change a setting, and a request to draft a paragraph, through a single input. The artificial distinction between “search for a thing” and “issue a command” — a distinction Raskin regarded as an accident of implementation rather than a fact about user intent — finally collapses. You type what you want; the system works out which kind of want it is. This is LEAP’s “one mechanism for everything” principle, arriving thirty-seven years late.
It is direct. The user expresses intent in the form closest to thought — language — rather than translating that intent into a sequence of menu selections and button clicks. Raskin valued directness because every translation step is a place for error and a cost in time. The prompt bar removes translation steps, at least nominally.
And it is invocable from anywhere, increasingly a single keystroke away regardless of what the user is doing. LEAP was always available; you did not enter a search mode and leave it. The best prompt bars approximate this, sitting one shortcut from any context rather than demanding that the user navigate to them.
These are real continuities. The prompt bar is, in a precise sense, the most LEAP-like thing the mainstream has produced.
What the Prompt Bar Gives Up
The inheritance is partial, and the parts it drops are exactly the ones Raskin cared about most.
LEAP was predictable. Holding the key and typing the same characters always moved you to the same place. The mechanism found what existed; it did not generate, interpret, or guess. The user could form a reliable mental model — type these characters, arrive there — and that model never betrayed them. The prompt bar, by contrast, interprets. The same request can yield different results depending on phrasing, on prior context, and on the system’s own changing behavior. Raskin’s interfaces were built to do what the user expected, every time, verifiably. An interpreting prompt bar trades that reliability for flexibility, and the trade is rarely acknowledged in the marketing.
LEAP was immediate and transparent. The cursor moved as you typed; you watched it happen; there was no hidden processing between your action and its visible effect. The prompt bar inserts an interpretation step that the user cannot see into. You type a request and receive a result, but the path from one to the other is opaque. When the result is wrong, there is no equivalent of releasing the LEAP key to find you had simply mistyped — there is only the request to try different words and hope.
And LEAP found what was there. A LEAP that matched nothing told you, reliably, that the thing was not present. A prompt bar that interprets and generates can produce a confident, fluent result that corresponds to nothing real. The failure modes are not symmetric: an empty search result is honest, while a fabricated answer is misleading, and the two can look identical at the moment of delivery.
The Lesson Raskin Would Draw
Raskin’s interest was never in novelty for its own sake; it was in whether an interface reduced the user’s burden and respected the user’s mental model. By that standard, the prompt bar is a genuine step toward the uniform, direct, intent-driven interface he wanted, and a genuine step away from the predictability that made his version trustworthy.
He would, I think, have welcomed the collapse of the search-versus-command distinction without reservation — that was his idea, and it took the industry far too long to accept it. He would have insisted, just as firmly, that the interpreting layer be held to account: that the system make clear when it is finding versus generating, that it expose enough of its reasoning for the user to form a working model, and that it never let fluency stand in for correctness. The point of LEAP was that you could trust it. A prompt bar that cannot yet be trusted in the same way is an unfinished version of the same idea, not a finished version of a different one.
A Direct Line, Not a Clean One
The honest summary is that the prompt bar of 2024 is LEAP’s heir, and an imperfect one. It has recovered the uniformity and directness that Raskin built into a machine the market ignored in 1987, and it has lost the predictability and transparency that made that machine worth defending. The lineage is real. Recognizing it is useful, because it tells you where to push: the work that remains is not to invent the conversational, intent-driven interface — Raskin already did that — but to give it back the trustworthiness it traded away on the road to the mainstream.
Related: