Apple Intelligence — The Year of the Flop
Apple Intelligence was supposed to be the iPhone moment for on-device AI. In May 2026, it's an embarrassment. The flagship Siri rebuild is delayed again, adoption is rounding to zero, and Apple's structural problem is now visible: they refuse to use third-party models, and their own are 18 months behind.
Apple Intelligence was supposed to be the iPhone moment for on-device AI. When Tim Cook unveiled it at WWDC 2024, the framing was: Apple has been quietly working on the right way to do AI, and they're now ready to ship something that competitors can't match. The vision included a fully rebuilt Siri that would understand context across apps, a suite of consumer AI features that worked privately on-device, and a partnership with OpenAI that would handle queries Apple's own models couldn't.
In May 2026, two years after that announcement, Apple Intelligence is an embarrassment. The flagship Siri rebuild has been delayed twice and still isn't shipping. The consumer features that did ship (Genmoji, Image Playground, Writing Tools) have adoption rates rounding to zero in independent surveys. The Private Cloud Compute infrastructure is technically working but barely used. And the competitive gap has widened, not narrowed — OpenAI, Anthropic, and Google have shipped meaningfully better AI experiences while Apple's offerings have stayed roughly flat since launch.
The honest assessment, after 24 months, is that Apple Intelligence is the largest strategic miss of Tim Cook's tenure as CEO.
What went wrong
Three structural problems compounded.
The Siri rebuild was overscoped. The promised Siri (context across apps, natural conversation, ability to perform multi-step actions) was a moonshot product. Apple's engineering teams reportedly underestimated the complexity of integrating LLM-style intelligence into the existing Siri infrastructure, which was built for a fundamentally different interaction model. The result has been delay after delay. The original iOS 18 target slipped to iOS 19, then to "early 2027" depending on the leak. Each delay reduces credibility with both developers and consumers.
The consumer features that shipped weren't valuable. Genmoji (custom emoji generation) is a novelty that most users tried once and never came back to. Image Playground (image generation) produces lower-quality output than competitors and has weaker UX than even basic AI image tools. Writing Tools (text rewriting, summarization) work but are duplicated by every productivity app users already have. None of the features solved a real problem users were asking to be solved. They were AI features for the sake of having AI features, which consumers see through quickly.
Apple's model quality is meaningfully behind. The on-device models Apple ships are reasonable for narrow tasks but uncompetitive with frontier models from OpenAI, Anthropic, and Google. When the user asks Siri something even slightly complex, the on-device model handles it poorly. Apple's solution is to route to ChatGPT through the partnership deal, but the handoff is rough, the latency is bad, and users notice they're being passed off to a third-party model that they could have used directly.
The three problems reinforce each other. The Siri delay means users have no reason to care about the consumer features. The weak consumer features give users no reason to wait patiently for the Siri rebuild. The model quality gap makes both the consumer features and the Siri experience worse than alternatives. The cycle is hard to break without fundamental change.
The structural problem nobody at Apple wants to face
Apple's product philosophy historically has been: own the full stack, ship when ready, prioritize user experience over feature count. This philosophy worked spectacularly for hardware, for the operating system, and for most software categories Apple has entered. It does not work for AI in the current moment.
The reason: AI capability is moving faster than Apple's "ship when ready" cadence can match. OpenAI, Anthropic, and Google ship significant model improvements every 3-6 months. Apple ships meaningful AI updates every 12-18 months at best. The gap compounds. By the time Apple ships a feature that uses their latest model, the competitive frontier has moved 2-3 generations ahead.
The right response is to acknowledge that being a frontier AI lab and being Apple are different jobs. Apple should focus on being the best platform for surfacing third-party AI capabilities to consumers — the equivalent of how Apple is the best platform for surfacing third-party music, third-party apps, third-party content. Not the source of the AI capability itself.
Newsletter
Like this decoded read? Get the next one in your inbox.
One sharp take on the latest in AI for PMs, founders and operators. Every few days. Free.
Free forever. Unsubscribe anytime. No spam.
This is a philosophical shift Apple's leadership has resisted. The Tim Cook / Craig Federighi position has consistently been that core experiences must use Apple's own technology. That position was correct for almost everything Apple has shipped historically. It's wrong for AI in 2026, and the cost of being wrong is growing every quarter.
What the Cluely-style partnership got right and wrong
The OpenAI partnership announced in 2024 was a half-step in the right direction. It acknowledged that Apple's models couldn't handle complex queries and that ChatGPT could. But the implementation was limited — ChatGPT can be called from Siri for specific query types, but only with explicit user permission per query, and the integration is rough. Users notice they're being handed off to a third-party model, the experience feels disjointed, and the partnership doesn't extend to the deeper iOS integration that would actually compete with what Google has built with Gemini.
The full version of the right move would be: let users pick their default LLM (ChatGPT, Claude, Gemini, Apple) the same way they pick their default browser or search engine. Make the integration first-class — the LLM has access to context across apps, can take actions on the user's behalf, can run in the background. Stop trying to be the LLM provider and start being the platform that surfaces the best LLM to each user.
This would be a meaningful philosophical concession for Apple. It would also probably save the AI strategy.
What's coming at WWDC 2026
The rumor mill suggests three plausible announcements.
Expanded third-party LLM access. The most-rumored move is opening Siri to Claude and Gemini as backend options, complementing the existing ChatGPT integration. This would be a meaningful acknowledgment that Apple can't catch up to frontier models alone and needs to leverage the competitive market in models.
Improved Apple Intelligence on-device models. Apple's silicon team has been working on dedicated AI inference hardware. WWDC 2026 may include a meaningful upgrade to Apple's on-device model quality, narrowing (but not closing) the gap with competitors. This would help with the privacy-positioned use cases but won't solve the fundamental capability gap.
App Intents expansion for AI integration. Apple has been quietly building infrastructure that lets third-party apps expose actions to AI agents (similar to MCP at a system level). WWDC 2026 may formalize and expand this, positioning iOS as a platform for AI agent integration rather than just a consumer of AI capability.
Any of these would help. None of them solve the underlying problem that Apple Intelligence as launched in 2024 was the wrong strategy. The honest move would be to retire the "Apple Intelligence" branding entirely and rebrand around what Apple is actually good at: building the best platform for great experiences, regardless of who provides the underlying capability.
What this means for the broader AI ecosystem
Three implications.
Big Tech distribution doesn't guarantee AI category wins. Apple has the most-coveted consumer distribution platform on Earth. They have not been able to translate that into AI category leadership. The lesson generalizes: distribution helps but doesn't substitute for capability. Companies racing to build AI features assuming "we have the users, the capability will follow" are missing what Apple has missed.
Privacy as positioning is necessary but insufficient. Apple's privacy story is genuinely strong and genuinely differentiated. It hasn't been enough to overcome the capability gap. Users care about privacy in surveys; they care about capability when they're actually using the product. The companies that win consumer AI will have both, not just one.
The model layer is more concentrated than people thought. Three years ago, the assumption was that many companies would build competitive frontier models. The reality in 2026 is that only OpenAI, Anthropic, and Google are at the frontier. Apple's struggle to build competitive models in-house, despite having Apple-level resources and engineering talent, is evidence that frontier AI capability is harder to develop than the early-2020s narrative suggested.
For Apple, the path forward requires philosophical change at the highest levels of the company. That kind of change is hard for organizations as successful as Apple has been. The likely scenario is incremental opening to third-party models over the next 2-3 years, with the Apple Intelligence branding quietly de-emphasized. Whether that's enough to recover the AI strategy or whether Apple permanently cedes the AI assistant category to other companies is the central strategic question for Apple over the next five years.
Frequently asked
Why is Apple Intelligence considered a flop?
Three reasons: (1) The flagship Siri rebuild promised for iOS 18 has been delayed twice and still isn't shipping in mid-2026. (2) The features that did ship (Genmoji, Image Playground, Writing Tools) have rounding-to-zero adoption among iPhone users. (3) Competitors have shipped meaningfully better AI experiences while Apple's offerings have stayed roughly flat since launch. The combination of unfulfilled promises and irrelevant features make 'flop' the consensus view among Apple analysts in 2026.
Why is Apple behind on AI?
Two structural reasons. (1) Apple's principle that core features must use Apple's own models has held back the consumer-facing AI experience because Apple's models are meaningfully behind frontier models from OpenAI, Anthropic, and Google. (2) Apple's privacy positioning, while genuinely strong, has constrained the kind of AI features they're willing to ship — they refuse training-data approaches that competitors use freely. The combination is principled but uncompetitive in 2026.
Will Apple open up Siri to third-party LLMs?
There are persistent rumors that WWDC 2026 will announce some form of third-party model integration, likely letting users choose ChatGPT or Claude as Siri's backend for certain queries. The Cluely-style deal between Apple and OpenAI from 2024 was a precursor but limited in scope. The broader opening to third-party models would be a significant philosophical shift for Apple but is increasingly seen as necessary to compete.
Is Private Cloud Compute working?
Technically yes, in usage no. Private Cloud Compute is Apple's encrypted cloud infrastructure for running AI on Apple's servers without Apple having access to the data. The engineering is reportedly working as designed. The problem is that the features running on PCC are not interesting enough for users to care. Solving a privacy problem nobody knew they had isn't a winning product strategy if the underlying capability is uncompetitive.
What should Apple do about AI?
The honest answer is open up. Allow ChatGPT, Claude, and Gemini as first-class Siri backends. Allow third-party AI providers in the App Store with deeper iOS integration than current API limits allow. Ship the on-device AI features that genuinely work (text summarization, image enhancement) without overselling. Stop trying to be a frontier AI lab and instead be the platform that surfaces the best frontier AI to consumers. The hardest part is the philosophical shift; the technical execution is straightforward once the shift is made.