PART01 A Make-Up Exam Costing $230 Billion

Apple CEO Tim Cook took the stage at Apple Park for the last time as CEO, delivering the keynote at WWDC 2026. Three months later, hardware engineering chief John Ternus will succeed him, leading Apple into a more challenging AI era.

This farewell show on June 8 was supposed to be a coronation. However, in reality, it looked more like a 'make-up exam.' The 'brand-new AI' promised at WWDC 2024 did arrive, now called 'Siri AI,' but from the live demonstration, it could answer questions, search for information, send messages, and set reminders—utterly unsurprising for users accustomed to AI tools. As for Agent capabilities, they were even further out of reach.

Following the event, Apple's stock price plummeted, evaporating over $230 billion in market value by the close compared to the day's peak.

Operationally, this was a 'make-up exam' barely passing.

Let's rewind the clock two years.

In June 2024, Apple boldly announced 'Apple Intelligence' at WWDC, promising that Siri would have capabilities such as screen content awareness, cross-app operations, and natural language understanding. It was Apple's first direct response to the AI wave sparked by ChatGPT, briefly exciting capital markets. Apple's stock price hit new highs that summer.

But promises quickly turned into delays.

In fall 2024, the new Siri failed to launch with iOS 18. In 2025, Apple postponed it again, citing 'internal technical issues.' Meanwhile, OpenAI released GPT-4o, Google introduced Gemini 1.5 Pro, and Anthropic's Claude 3 surpassed in coding and reasoning. Apple's intelligent assistant often set alarms incorrectly.

More embarrassingly, internal data revealed that in March 2025, Apple executives admitted during an internal meeting that Siri's new features had a success rate of only 60% to 80%. This meant one failure every five interactions. When this figure circulated within the tech industry, it was described as 'ugly and embarrassing.'

In January 2026, Apple reached a cooperation agreement with Google, integrating the Gemini model into Siri—equivalent to a top student having to borrow notes from a deskmate.

When Apple Senior Vice President Craig Federighi stated at WWDC 2026, 'We believe privacy in AI is non-negotiable,' he had to simultaneously explain why your Siri queries would be sent to Google's servers for processing. Apple gave this compromise a nice name: Private Cloud Compute. Essentially, this was Apple admitting that its self-developed AI model could not independently support the capabilities expected of a modern intelligent assistant.

Additionally, the list of new features showcased at WWDC 2026 was not particularly novel for many users.

Siri now has a standalone app, supporting both text and voice interaction modes while retaining conversation history. ChatGPT had already achieved this in 2022. 'Visual Intelligence' integrates with the camera to recognize objects, scan business cards, and read nutrition labels. Google Lens had already launched in 2017.

Image Playground has been upgraded to support higher-quality image generation and 'description modification' features, which users of Midjourney and Stable Diffusion have been enjoying for three years.

As for Safari incorporating AI tag management, Messages offering AI reply suggestions, and the Phone app reading email context during calls, these features have existed on Android and third-party apps for years. Furthermore, Apple admitted that some Apple Intelligence features (such as image generation) have usage restrictions. Due to regulatory requirements, users in the EU and China are excluded from this 'AI upgrade.'

PART02 The Cost of Closed Philosophy

After WWDC 2026, tech media evaluations were surprisingly consistent.

Mashable recorded a detail during its live coverage: when Apple Senior Vice President of Software Engineering Craig Federighi announced that Siri AI would launch as a 'public beta' later that year, the top-rated comment in the chat was: So... that's it? More pointed criticism came from Bloomberg's Mark Gurman. Before the event, he posted on X, predicting that Apple would focus on discussing 'privacy and security features,' implying not to expect genuine AI breakthroughs.

These reactions point to a core issue: Apple has lost its ability to 'define standards' in the AI era.

That was its core competitiveness in the smartphone era. Over the past two decades, Apple defined what smartphones (iPhone), tablets (iPad), and smartwatches (Apple Watch) should be. It didn't need to be the first to market, but it always redefined industry standards.

However, in the AI field, it has been playing a follower role. Reflected in the capital market, despite its core businesses (iPhone, Mac, services) performing strongly, its stock price performance in 2025 was lackluster. Entering 2026, as investor enthusiasm clearly shifted toward companies already generating clear AI revenue, Apple's crisis became more apparent.

An Apple engineer demonstrates AI's ability to adjust and generate new photos on stage

Apple's foundations are undeniably solid. Beyond technical capabilities and cash reserves, its most competitive asset is owning the world's largest device ecosystem—no AI company can directly reach 2.35 billion active devices globally. Additionally, Apple's privacy commitments are highly recognized among users, effectively alleviating many users' concerns about AI security.

However, this is a double-edged sword.

The elegant product logic of closed systems was Apple's ace in the smartphone era. From Macintosh to iPhone, from iOS to App Store, Apple created an ultimate user experience by controlling every aspect of hardware, software, and services.

This closed nature was a moat in the mobile internet era but has become a cage in the AI era. The competitiveness of AI models is directly tied to data scale, yet Apple's strict privacy policies severely limit internal teams' access to user data. Researchers can only train models using third-party authorized datasets and synthetic data—akin to making an adolescent drink only water and eat buns daily, inevitably leading to malnutrition and stunted growth.

When OpenAI released ChatGPT in 2022, it didn't need to consider device compatibility, regional differences in privacy policies, or maintain a local computing architecture across billions of devices. It only needed a browser to let users worldwide experience cutting-edge AI capabilities. What about Apple? An insider described it as: Doing AI at Apple means every move faces hundreds of rejections. You have to fight the 'privacy police' to make progress.

Under such circumstances, speed inevitably becomes a sacrifice.

While competitors iterate models weekly, Apple advances features annually. The term 'AI Agent' has been hot for a year, yet Apple continued to 'tease' at WWDC 2026. It allows users to book restaurants, manage schedules, and control smart homes via Siri, but most Agent functions are still 'scheduled to launch.'

PART03 Changes at the Transition

But changes are happening. The keyword is openness.

First, Apple has humbled itself to make friends.

The latest iOS version allows users to freely switch to third-party AI models approved by Apple for the first time. From Google Gemini and Anthropic Claude to DeepSeek, all can replace Apple's self-developed model to power Siri, writing tools, and image generation.

Apple's relationship with Google has become even tighter. At WWDC 2026, Apple officially announced: To enable Siri to handle the most complex requests, we are collaborating with industry-leading AI models. Today, we are pleased to announce that Google Gemini will be deeply integrated into Apple Intelligence.

The two companies, mortal enemies in the smartphone market and litigation opponents in search and advertising, were forced to ally in the AI era. This resembles Microsoft's $150 million investment in Apple in the 2000s to sustain Office for Mac.

Macworld 1997, Jobs on stage, Bill Gates announcing investment + Office 98 for Mac via remote video

Face means nothing in the face of circumstances.

Second, Apple is attempting to build a more open developer ecosystem.

Specific measures include: Siri Extensions API allows third-party AI models to access system-level experiences, Core AI framework replaces the outdated Core ML, Foundation Models v2 enables calling on-device large models with just three lines of Swift code, and establishing a dedicated App Store AI section.

Clearly, Apple is trying to replicate its 2008 playbook. Back then, in the smartphone blue ocean, developers who learned Objective-C and Apple's Human Interface Guidelines could reach the world's most affluent user base. In the App Store's first year, developer payouts exceeded $1 billion; by 2017, this figure had grown to $70 billion.

Using an open ecosystem to compensate for its lack of innovation capabilities, letting developers explore application scenarios Apple hasn't yet considered. This was Apple's path in the previous era, and now it hopes to replicate it in AI.

But the difficulty remains high. From all angles, this move clearly differs from the proactive innovation of 2008 and instead carries a strong whiff of forced compromise.

For example, Apple is trying to control the AI entry layer. According to information revealed at WWDC 2026, App Intents becomes the sole interface for third-party apps to be invoked by Siri, while SiriKit is officially retired. Apps that haven't migrated will become invisible to AI. However, the current core competitiveness of AI lies not in entry points but in model capabilities, data flywheels, and iteration speed. When developers can directly call Google Gemini or OpenAI via Siri Extensions, Apple controls only the switch, not the power grid.

Once users perceive that third-party AI experiences significantly outperform Apple's self-developed ones, they will bypass Siri to use native apps directly—rendering Apple's entry-point 'teasing' obsolete and reducing it to an expensive middleman.

Apple's reliance on Google Gemini carries similar risks: If Siri's core intelligence comes from Google, why wouldn't users just use Google Assistant or the Gemini App directly?

As Cook steps down, the answers to these questions await new CEO John Ternus. He has a hardware engineering background and led the development of the M-series chips and Vision Pro. The choice of this successor itself signals that Apple still believes the ultimate competitiveness of AI lies in the integration of 'chips + devices + ecosystem.'

Cook completed his tenure with an 'AI make-up exam,' and Apple has abandoned its arrogance of developing everything in-house, beginning to participate in AI competition more pragmatically. Collaboration with Google, openness to third-party AI models, and embrace of the Agent concept—these shifts, while late, at least seem to be heading in the right direction.

John Ternus still has the opportunity to find a new version of Apple's signature 'late-mover advantage'—of course, all contingent on Apple maintaining its hardware advantages.