Apple WWDC: Will Tim Cook Deliver a Final Surprise Before Stepping Down?

'Tim Apple' says goodbye to the big stage today

On Monday, June 8, Apple kicks off its Worldwide Developers Conference, and it will be Tim Cook's last major appearance. In September, he will transition to the role of Executive Chairman, and John Ternus, previously head of hardware, will become Apple's third CEO after Steve Jobs and Cook. In terms of content, everything revolves around an AI reboot: According to The Information, Apple intends to sell its on-device AI processing as a privacy feature, while Google, Meta, and OpenAI are investing billions in cloud data centers. To achieve this, Apple is training a smaller model using a large version of Google's Gemini, which will run locally on the iPhone, Watch, and Mac, and is apparently acquiring startups that specialize in this exact shrinking process. Siri is finally getting the overhaul that has been expected for two years and, on the screenshots reconstructed by Bloomberg, looks like Claude, ChatGPT, or Gemini, just in Dark Mode. In addition, Apple has approved Poke, the first AI agent allowed to contact users via the iMessage for Business platform. → Tech Brew

Synthszr Take: Apple is turning its biggest technical limitation into its strongest selling point, and that's clever. While others are building their moats out of data centers, Apple is sitting on a billion devices with its own chips that can process requests locally. This is the same logic I described back in June 2024 regarding the DMA dispute: the more closed the system, the fewer attack vectors, and privacy is technically difficult to reconcile with radical openness. The fact that Apple is shrinking a Gemini model, of all things, to become independent of the cloud has a nice irony to it, but this has always been Apple's formula: not to be first, but to set the tone once others have prepared the party. Ternus is inheriting a Siri that's two years behind schedule and now looks like the competition, and that's the real litmus test. Velocity is the solution to almost all software problems, and that's exactly what Apple Intelligence has been lacking so far. We'll see if the new man can find that pace starting today.

Apple Approves the First AI Agent in Messages for Business

Poke, developed by The Interaction Company of California in Palo Alto, is the first standalone AI agent that Apple is allowing on its Messages for Business platform. Previously, this channel was reserved for airlines, retailers, and hotel chains to communicate with their own customers via iMessage. Poke makes AI agents as simple as an SMS and already runs via SMS, Telegram, and in some markets, WhatsApp; now iMessage is being added. Since its launch in March, the service has handled around 100 million messages, assisting with daily planning, calendars, health tracking, and smart homes. The business model is interesting for founders: Co-founder Marvin von Hagen pays Apple per user, significantly less than with Meta AI, which raised its fees for third-party agents on WhatsApp following EU regulations. The news comes days before WWDC 2026, where Apple is expected to showcase an AI-optimized Siri and possibly open the App Store to agents. → The Deep View

Synthszr Take: Apple is currently building a toll booth for the Intent Economy, and the currency is a per-user fee. iMessage is pre-installed on hundreds of millions of iPhones—no download, no onboarding, no friction: whoever is here sits in the most valuable inbox in the world. This is precisely the new distribution price that every agent startup must now factor in, while the 'build moat' continues to dry up. In March, Apple opened Siri to ChatGPT and others; now, the business channel is open to third-party agents. The pattern is clear: Apple isn't selling the model, but access to the touchpoint. The fact that Poke gets a much cheaper rate than at Meta shows how early and contested this toll structure still is, and that's exactly why it's worth getting in now while the fees are low and the spots are open. Anyone building an agent should treat distribution through pre-installed messaging channels as the primary lever, not just a bonus. Today's WWDC will reveal whether Apple is throwing this door wide open or just leaving it ajar.

How Apple's AI Course Radically Shifted in 2025

In his Power On newsletter, Mark Gurman reconstructs a secret meeting in early 2025 that turned Apple's AI course around. In a conference room near Craig Federighi's software division, the SVPs, COO, and CFO gathered, convened by the now-retired Jeff Williams. Cook himself was not present but had already lost faith in AI chief John Giannandrea. The topic was the crisis: Apple Intelligence was flopping, the Siri overhaul was facing a delay, while Meta, Microsoft, Google, OpenAI, and Anthropic were pulling ahead. Mike Rockwell, the man behind the Vision Pro, volunteered to fix Siri but clashed with Federighi, who didn't want AI to be moved out of his software organization. Giannandrea was stripped of his power and left, and former Google and Microsoft executive Amar Subramanya came in as the second AI head. On Monday at WWDC, Apple will show the result: a new Siri whose models and cloud infrastructure will now run on Google's Gemini and Google Cloud. → www.bloomberg.com

Synthszr Take: A decade ago, hardware chief Dan Riccio warned internally that AI could become an existential threat to Apple's devices. At the time, the leadership team wouldn't listen; Rockwell was never allowed to implement his five-year roadmap for Siri. This is precisely Apple's most expensive weakness: the company recognizes tectonic shifts years in advance and then gets blocked by its own internal turf wars—in this case, the dispute over whether Rockwell should report to Cook or Federighi. The corporation's immune system resisted the disruption longer than it took the competition to pull ahead. The fact that Siri now runs on Gemini and Google Cloud is an honest admission that its own models were below the industry average (we saw this coming in February with the Siri disaster and in April with the Cook transition). Apple has enough money and distribution power to buy expertise, and the move is the right one. But the real question will be asked on Monday and thereafter: Can a corporation that has treated AI as a software feature for years learn to see it as a core quickly enough, especially when that core now comes from a competitor?

The AI Bet Is Absorbing the Entire Capital Market

The figures from the Kobeissi Letter are hard to fathom: AI-related companies have issued around $140 billion in investment-grade bonds this year. That accounts for 49 percent of all IG issuance on the market. In venture capital, the concentration is even more extreme: about $220 billion flowed into AI firms, a full 87 percent of the total VC volume. Practically every second dollar in the market for top-tier corporate bonds now has an AI label. And in venture capital, there's hardly anything left for anything that isn't related to large language models or data centers. The data comes from a market overview by Ed Sim. → Ed Sim from What's Hot 🔥 in Enterprise IT/VC

Synthszr Take: 87 percent of all venture capital in a single thematic group—this is unprecedented. What makes me nervous isn't the VC. Risk capital is allowed to bet; that's what it's for. It's the $140 billion in investment-grade bonds that worries me, because this is where pension funds and conservative funds, which are supposed to be buying security, are getting involved. They are currently financing data centers and GPU orders whose depreciation cycles no one really knows. In May, SoftBank already restricted loans to OpenAI; that was the first crack in the structure. When a single sector makes up half of an entire capital market segment, it's not a sign of strength, but of a lack of diversification; and historically, a lack of diversification has always been the point where things get expensive. Anyone buying bonds now should understand that they are making a technology bet, even if the rating claims otherwise.

Nvidia: Every Edge Device Will Become Autonomous

On a public stage, Nvidia CEO Jensen Huang has sharpened his thesis: Every device at the edge of the network, from sensors to factory machines to vehicles, will, sooner or later, act autonomously. With this, Huang extends the AI narrative beyond the data centers where Nvidia generates the majority of its business today. Edge computing here means inference directly on the device, without a constant detour through the cloud. For Nvidia, this is the logical extension of its platform: first the GPUs in the training data center, then the chips and software stacks in every autonomous machine. Huang is therefore not just selling silicon, but also an end-to-end operating system for the physical world. He did not provide a concrete roadmap with timelines on stage. → Tom's Hardware

Synthszr Take: Huang has been selling the same movement for months, just taking it one level further each time. In January, it was the Android gambit—free software that locks customers in forever; in March, he wanted the whole world; now, the world is also supposed to become autonomous, and on his stack. This is clever architecture: Whoever owns Layer 1 (the infrastructure) and gives away the software all the way to the edge device is building a moat that's nearly impossible to leave. The honest question is not if edge inference is coming (it is, because latency and privacy will force it), but whether a single provider should be allowed to control the entire chain from the factory floor to the token broker. Anyone setting up pilot projects today should keep the model layer interchangeable and not cede their own orchestration logic to Nvidia's proprietary conveniences. The vision is probably right. The price for this is lock-in, and it will get more expensive in 18 months.

Meta's Manus Integrates Shopify Directly into Chat

Manus, the AI agent startup that Meta acquired for around $2 billion in late 2025, is now integrating Shopify directly into its chat. The idea: You describe your business in a prompt, and Manus builds the storefront, connects it to Shopify, and sets up a functional shopping cart. Branding, product pages, navigation, checkout—everything is created in the same conversation, in a Shopify development store under your name. Only when you click 'Claim' and log in to Shopify does the store go live. In the background, Shopify continues to handle hosting, payments, taxes, and shipping, while Manus takes care of the daily work on top: Excel spreadsheets with SKUs become product lists, sales data turns into campaigns, and weak products become discount promotions with a ready-made email draft. Those who already run a Shopify store no longer need to coordinate across multiple tools but can work directly in the chat with their existing store. → Techpresso

Synthszr Take: The tool layer is becoming a commodity right before our eyes. Building a storefront, writing product descriptions, choosing a theme, matching images to SKUs: for years, this was the work of agencies and freelancers, and Manus does it in a single conversation. What's interesting is where the moat now lies. Shopify retains payments, taxes, and shipping—the regulated, inconvenient infrastructure that can't be chatted away. The real competition is shifting to the pipeline logic behind it: Which brand has its brand voice, product knowledge, and campaign hypotheses so well-organized that an agent can use it to drive real sales figures instead of producing interchangeable content? The fact that Meta is behind Manus makes it tangible. The corporation that owns Instagram and the ads is currently building an end-to-end path from the ad to the checkout. Anyone opening a store today should try out the development store before signing the next agency invoice.

OpenAI Is Building an Emergency Exit for Prompt Injection

OpenAI is rolling out 'Lockdown Mode,' an optional security feature designed to protect against prompt injection attacks. In this method, attackers hide malicious instructions on websites or in documents that the model reads while researching and then executes. When the mode is activated, users lose functionality: ChatGPT no longer pulls images from the web, downloads files for analysis, and Deep Research and Agent Mode are completely disabled. OpenAI itself says the mode is not for everyone, but for people and organizations that handle sensitive data and want to adhere to stricter guardrails against data leakage. The protection doesn't prevent injections from occurring in the processed content but limits the network requests through which someone could extract data from the account. This is available for all personal accounts, including the free tier. In parallel, a session manager is being introduced, allowing users to view active devices and log out of them individually or all at once (though this can take up to 30 minutes). → Techpresso

Synthszr Take: The most honest sentence in the entire rollout is: 'Most users don't need to use the feature.' Translated, this means: The agent features that OpenAI has been selling as the future for months (Deep Research, Agent Mode, autonomous browsing) are the attack vectors, and the only clean protection is to turn them off. This is the structural weakness of any system that autonomously reads from the open web while also having access to your data: it cannot reliably distinguish between an instruction you give it and a hidden one on a third-party site. Lockdown Mode doesn't solve this; it merely shifts the responsibility to you, with the caveat that you sacrifice half the functionality. For the banks, law firms, and hospitals that are supposed to use OpenAI's business plans, this is an inconvenient truth, especially now that growth targets are reportedly shaky according to April reports. Anyone connecting agents to real business processes today should be aware of this switch and know exactly what it costs. Building security as an opt-out instead of a default is a bet that the first major data breach won't bear your own name.

Vibe Coding: How AI-Powered Development Is Revolutionizing the Workplace

A mother with no programming experience builds a baby nutrition app in five days. An artist develops a private jet tracker as an early warning system for the apocalypse. A student launches a vintage marketplace during her final semester. What these stories have in common is that they all use 'Vibe Coding'—the new way of developing software, where AI models like Claude or ChatGPT write the code while humans provide the ideas. Business Insider is documenting this movement that is currently turning the tech industry upside down. Lisa Lin, who worked in the tech industry for years but had no coding experience, needed a special nutrition app after the birth of her child. Instead of spending months learning to code or hiring developers, she described her vision to Claude—and had a working app within days. Kyle McDonald took it a step further: he used Vibe Coding to track the movements of over 35,000 private jets as a kind of early warning system for social upheavals (if the super-rich suddenly all fly in one direction, something might be wrong). Hana Elster built a functioning vintage marketplace in just five days during her last semester of college—as a side project to her studies. → Business Insider

Synthszr Take: The numbers speak for themselves: where months of development time and six-figure budgets were once required, five days and a clear product vision are now enough. This is the real revolution behind 'Vibe Coding'—the bottleneck is shifting from technical know-how to the quality of the idea. 32-year-old Lisa Lin proves it: if you want to solve a real problem and can articulate it precisely, you no longer need ten years of coding experience. The democratization of software development is already underway while we are still debating AI regulation. What will happen to the 28 million professional developers worldwide when suddenly anyone can code? Probably the same thing that happened to photographers on Instagram: the best will get better, the average will become replaceable. The real skill in the future will lie in 'intent'—the ability to ask the right questions and develop products that people truly need.

More People Are Now Appearing in Court Without a Lawyer, Thanks to ChatGPT

The number of lawsuits filed by US citizens without a lawyer has risen from 11 percent in 2022 to 16.8 percent in 2025, and within these cases, the volume has more than doubled compared to the period before 2023. A study by Anand Shah (MIT) and Joshua Levy (USC) analyzed 4.5 million civil cases and ran 1,600 documents through the AI detector Pangram: the proportion of AI-generated texts climbed from 1 percent (2023) to 18 percent (2026). Judge Maritza Braswell in Colorado recognizes the models by their style, hallucinated judgments, and fabricated citations, but says the briefs are often better formulated and quicker for her to process. In Vermont, where a viral Reddit post explained step-by-step how to draft an application using Microsoft Copilot, self-filed cases soared from about 45 per year to over 1,100 in 2024. However, the chances of winning are not improving: those who file without a lawyer continue to lose significantly more often. In parallel, courts are grappling with new questions, such as whether chatbot conversations enjoy the same protection as attorney-client privilege, with conflicting rulings from Michigan and New York in February. → The Download from MIT Technology Review

Synthszr Take: This is the Jevons paradox in court: the cost of drafting a lawsuit drops to zero, so more lawsuits are filed. This is the same amplifying logic I know from the coding world. AI makes those who already had access to justice faster, and it gives a voice to those who could previously only submit illegible notes. What's interesting is that the judiciary benefits: Braswell openly says she can help better when she understands the argument. But the success rate remains unchanged because lawsuits are not just about text production, but also about strategy, evidence, and procedural rules, and that's precisely where the language model fails. The uncomfortable question is about liability: if Claude sends a client to defeat with a fabricated precedent, the user bears the risk alone, while the New York ruling makes it clear that these chats can even be used against them. Anyone selling AI as legal advice without addressing this asymmetry is building access to justice on sand.