June 26, 2026
At WWDC 2026, Apple announced what it is calling Siri AI: a rebuilt assistant with onscreen awareness, access to personal context across mail, messages, and calendars, and the ability to execute multi-step tasks across applications from a single spoken or typed request. iOS 27 ships with this as its central intelligence layer, powered by an architecture that Apple describes as combining on-device processing, Private Cloud Compute, and Google Gemini models for broader-knowledge queries.
The announcement uses language that Jef Raskin would have recognized. “Onscreen awareness” is close to what Raskin called context-sensitive response to the visible state. “Cross-app actions from a single request” is adjacent to what LEAP offered in 1987: a single gesture mechanism that worked identically regardless of what content it touched. The temptation is to read iOS 27 as a vindication of the design philosophy Raskin spent his career arguing for. That reading is partly right and partly a category error worth unpacking.
What Raskin’s Modeless Principle Actually Required
Raskin’s argument against modes was precise. A mode is a system state in which the same user input produces a different result than it would in another state. The problem is not inconvenience; it is the cognitive burden of tracking hidden state. When a user forgets which mode the system is in — which application is active, which tool is selected, which dialog is waiting for input — and acts on a false assumption, the interface has failed them. The error is not user error. It is a predictable consequence of a design that requires users to maintain an accurate model of system state in order to predict outcomes.
LEAP, the mechanism Raskin built into the Canon Cat and developed further in Archy, was the practical implementation of this principle. It was not a better search box. It was a single interaction model that applied uniformly to everything: navigation, search, command execution, and text editing all operated through the same physical gesture — hold the LEAP key, type characters, release when the cursor arrives. The uniformity was the design. Because LEAP worked identically everywhere, users did not have to track which state they were in. Any state produced the same result from the same action.
Where iOS 27 Aligns
The Siri AI architecture announced at WWDC 2026 has several properties that move genuinely in Raskin’s direction.
The onscreen awareness feature is notable. Siri AI can observe what is visible and answer questions or take actions relevant to that content, without requiring the user to navigate to a separate application or invoke a different interface for different content types. This reduces — though does not eliminate — the mode-switching burden. A user reading a document who wants to act on it can address Siri without first establishing which application context they are operating in.
The cross-app actions capability is the stronger alignment. Apple’s architecture, using App Intents as the underlying mechanism, allows a single request to span multiple applications: a user can ask Siri to find a flight confirmation in email, create a calendar event from it, and share the event details with a contact — without manually switching between Mail, Calendar, and Messages. The user gives a single instruction and the system manages the application transitions. At the level of user experience, this looks like modelessness: the applications become infrastructure the user does not have to manage.
The personal context system — Siri drawing on mail, messages, photos, and calendar without requiring the user to specify which data source to search — also fits the spirit of LEAP. Raskin’s universal search vision was that finding anything should require the same gesture, with the system responsible for determining what “anything” meant. A query that traverses multiple data sources without requiring the user to route the query manually is closer to that ideal than the pre-2024 world of separate per-app search boxes.
The Persistent Gap
The alignment is real but incomplete, and the gap matters.
The Siri AI interface is not itself modeless. To invoke it, users switch into a distinct Siri interaction state — a different mode with its own conventions, its own visual presentation, and its own set of expected inputs. To exit it, they return to the application surface they were on. LEAP did not work this way. There was no “LEAP mode” to enter and exit. The LEAP key was a quasimode: it was active only while physically held, and it produced its effect continuously as long as it was held. No separate interface appeared. No conversational turn-taking was required. The user remained in continuous contact with their content rather than delegating to an assistant and waiting.
The conversational model itself introduces a structural mode. A dialogue has turns; during the AI’s turn, the user waits. During the user’s turn, the AI waits. The system transitions between these states in ways the user does not control. Raskin was skeptical of conversational interfaces precisely because they impose this turn structure, which is itself a mode. His objection was not to voice input but to the sequential-negotiation model it typically requires.
There is also the question of habituation. Raskin argued that a humane interface should allow users to form habits — to reach a point where correct operation is automatic rather than deliberate. Conversational AI resists habituation because its responses are not fully predictable. The exact phrase that triggered a useful Siri action last week may not trigger the same action this week, because the underlying model has updated, or because context differs. LEAP’s behavior was deterministic: the same physical action on the same content produced the same result. Users could automate their interaction through practice. Conversational AI keeps users in a more deliberate, attentive mode of operation.
The Market Gap Raskin Identified
It is worth being fair about what Apple is attempting and why the conversational model is not simply a failure of nerve.
The modeless ideal requires a computing environment designed from the ground up to support it. Archy required that model; the Canon Cat required it. Neither could be grafted onto an existing ecosystem of independently developed applications with their own interfaces and conventions. Apple’s App Intents architecture is an attempt to standardize the capability layer beneath those applications without requiring their redesign, which is a genuine engineering achievement. It extends the reach of a single request without rewriting the applications that receive it.
What the App Intents architecture cannot do is eliminate the modes inside those applications. Once Siri AI completes a cross-app action and returns control to the user, the user is back in whichever application state the action ended in. The modes of the destination application are unchanged. The user still needs to track where they are.
Raskin would have called this a useful reduction in switching cost rather than the removal of switching. The distinction matters to the framework, even if it is less visible to users experiencing the convenience of the new system.
What the Comparison Clarifies
The comparison to Raskin’s work is useful not because it finds Apple insufficient — that would be an unfair standard — but because it sharpens what the modeless promise actually requires in order to be kept.
iOS 27’s Siri AI is the most significant step Apple has taken toward a single-gesture, cross-system interaction model since Spotlight launched in 2004. The onscreen awareness and personal context features specifically address the mode-tracking burden at the level of data access, which is where most users experience it most acutely. These are not cosmetic improvements.
What has not changed is the interaction model itself. Users still negotiate turns with the assistant, still return to modal application surfaces after each AI action, and still cannot form automatic habits around an interface whose outputs are inherently probabilistic. Raskin’s framework would describe this as a system that reduces the cost of entering and exiting modes without reducing the modes themselves.
Whether that distinction matters in practice depends on what you think the real problem of modern computing is. If it is the friction of getting things done across fragmented application ecosystems, iOS 27 addresses it directly and well. If it is the cognitive burden of tracking hidden system state, the problem remains — and WWDC 2026, like every WWDC before it, did not fully attend to it.
Frequently Asked Questions
What did Jef Raskin mean by a “mode” in interface design?
Raskin defined a mode as any system state in which the same user input produces a different result than it would in another state. The danger isn’t complexity but hidden state: when users lose track of which mode they’re in and act on a wrong assumption, the resulting error is a predictable design failure, not user error.
Does Apple’s new Siri AI eliminate modes the way Raskin’s LEAP did?
No. Siri AI reduces mode-switching costs through onscreen awareness and cross-app actions, but invoking it still means entering a distinct Siri interaction state with its own conventions, then returning to the prior application afterward. LEAP had no separate mode to enter or exit — it worked identically, continuously, wherever it was held.
What is Apple’s App Intents framework, and how does it relate to modelessness?
App Intents is the architecture letting a single Siri request span multiple applications — finding an email, creating a calendar event, sharing details — without the user manually switching apps. It reduces switching cost between applications but leaves each destination application’s own internal modes fully intact once control returns to the user.
Why was Raskin skeptical of conversational interfaces?
Conversational interaction imposes a turn structure: the system waits during the user’s turn, the user waits during the system’s. That turn-taking is itself a mode. Raskin also doubted users could habituate to conversational AI, since a probabilistic assistant may not respond identically to the same phrasing twice, unlike LEAP’s fully deterministic behavior.
Is iOS 27’s Siri AI a genuine step toward Raskin’s modeless vision?
Partially. Its onscreen awareness and personal-context search address the mode-tracking burden at the level of data access, which is real progress. But the underlying interaction model — a distinct assistant mode entered and exited, with turn-based dialogue — is unchanged. Raskin would call this reduced switching cost, not removed switching.
Related reading in this archive: