In partnership with
In Today’s Issue:
🍎 Apple’s Siri AI reset finally ships, just not on EU iPhones
🇪🇺 Apple blames the Digital Markets Act for the EU delay
💻 Google’s Gemini slips into Apple’s developer tools
🤖 Figure’s humanoids pull a real warehouse shift
✨ And more AI goodness…
⚡ The Signal
Today’s issue starts at WWDC, where Apple finally shipped the Siri reset it has promised for years. It gets a dedicated app, personal-context search, screen awareness, web answers, and systemwide app actions. The harder question is who gets to use it, and where.
Apple is blocking Siri AI on EU iPhones at launch and blaming the Digital Markets Act, while Google quietly slips Gemini into Apple’s own developer tools. The contest is moving past model quality into the ground that actually decides adoption: operating-system access, developer hooks, regulation, and whether an assistant can touch private data without becoming a liability.
All the best,
Kim Isenberg
🇪🇺 Apple delays Siri AI on EU iPhones
Apple says Siri AI will not be available on iOS 27 or iPadOS 27 in the EU at launch, blaming the Digital Markets Act and failed negotiations with regulators. The company says Mac and Vision Pro users in the EU will still get Siri AI, while iPhone, iPad, and watchOS access remain blocked without a timeline.
👉 tl;dr: Apple is turning Siri AI into a privacy-and-access fight with Brussels, not just a product delay.
🧭 OpenAI sketches its third phase
OpenAI says it is entering a “third phase.” Most of the post is broad-benefit language about making advanced AI abundant, affordable, and widely available. The interesting aspect however sits underneath it: OpenAI now expects AI-assisted AI research to become a determining factor in how fast progress comes over the next few years.
👉 tl;dr: OpenAI is pairing a broad-benefit narrative with a much more explicit claim about automated AI research.
💻 Gemini moves into Apple developer tools
Google says Apple developers can now call cloud-hosted Gemini models through Apple’s Foundation Models framework, starting with a preview release after WWDC. Gemini is also being integrated into Xcode for multi-step development tasks without switching tools.
👉 tl;dr: Apple’s AI developer layer is opening just enough room for Google to become part of the Apple app stack.
Turn Apple’s keynote into a filter for hype
Why it helps: WWDC mixes features that ship to everyone with ones gated by device, region, or developer support, and the keynote blurs the difference. Sorting them shows what you can actually rely on now.
Try this: Paste Apple’s WWDC feature list and ask: “Sort these into three buckets: shipping broadly now, gated by device or region, and dependent on developer adoption. Then write the product-risk memo Apple would not put on a keynote slide.”
🎬 Watch This
Apple WWDC 2026 June 8: Introducing Siri AI and more
Apple’s WWDC video lays out the company’s iOS 27 software stack, with Siri AI as the centerpiece: personal-context search, screen awareness, app actions, a dedicated Siri app, and Apple Intelligence features across the system. The useful watch point is how many of these assistant capabilities depend on device eligibility, region, language, and developer integration.
– Craig Federighi, Apple SVP of Software Engineering, presenting Siri AI at the WWDC 2026 keynote
The specific sticking point is a proposal Apple calls its Trusted System Agent. Apple says EU regulators rejected it, and that the Digital Markets Act would otherwise force the company to give rival assistants the same deep device access on a timeline it considers unsafe. That is Apple’s framing, and it is self-serving. But the underlying tension is real: assistant interoperability is easy to write into policy and hard to make safe once agents can send messages, buy things, open files, and change account settings.
Apple’s Siri AI reset finally has product shape
The Takeaway
👉 Siri AI gets personal-context search across messages, emails, photos, and more.
👉 Its on-device brain is a new 20-billion-parameter model (AFM 3 Core Advanced) that activates just 1 to 4 billion parameters per request, kept in flash to fit on iPhone.
👉 It can answer questions about on-screen content, use web knowledge, and trigger systemwide app actions.
👉 Apple says the features start developer testing now, with user beta access later this year for supported English-language devices.
👉 The EU rollout is split: Mac and Vision Pro get Siri AI, while iPhone and iPad do not at launch.
Apple used WWDC26 to show the Siri overhaul it had to show. Siri AI is pitched as a new version of Siri, not a small upgrade: it gets a dedicated app, conversation history synced through iCloud, personal-context search, on-screen awareness, broad web answers, and app actions that can operate across the system.
The technical story is more interesting than the keynote let on. Apple’s third-generation Foundation Models are a family of five built with Google, split between on-device models and larger ones on Private Cloud Compute. The standout runs locally: AFM 3 Core Advanced is a 20-billion-parameter model on the iPhone itself! To fit, it keeps the full model in flash storage and loads only 1 to 4 billion parameters into memory per prompt, selecting a small set of experts for each request rather than swapping weights token by token. That sparse, flash-resident design is what lets a 20B model run on a phone at all, and it is what powers the new on-device expressive voices and higher-accuracy dictation.
Google’s WWDC-week announcement makes the platform shift even clearer. Gemini can now sit behind Apple’s Foundation Models framework through Firebase, and Gemini in Xcode gives developers an agentic coding workflow inside Apple’s own toolchain. Apple is trying to keep the user-facing assistant native. The developer layer is already more porous.
Why it matters: The near-term test is whether Apple can make Siri AI useful without making it reckless. The longer-term test is whether regulators accept Apple’s privacy argument once assistant-level AI becomes a core OS capability.
Sources:
🔗 https://www.apple.com/newsroom/2026/06/apple-unveils-next-generation-of-apple-intelligence-siri-ai-and-more/
🔗 https://www.apple.com/newsroom/2026/06/due-to-dma-siri-ai-delayed-in-eu-for-ios-27-and-ipados-27/
🔗 https://blog.google/innovation-and-ai/technology/developers-tools/bringing-gemini-models-to-apple-developers/
🔗 https://machinelearning.apple.com/research/introducing-third-generation-of-apple-foundation-models
TARS DexHand: Giving Robots the Sense of Touch
At ICRA 2026 in Vienna, embodied AI startup TARS is debuting DexHand, a 21-DoF hand with tactile sensors detecting textures down to 0.05mm. Paired with the AWE 3.0 foundation model—which includes a newly released technology called TacForeSight that allows the robot to predict contact events before they fully occur—the system achieves "hand-brain integration." Live: their A1 robot packs a backpack and performs sub-millimeter wire harness assembly with real-time error correction by using the same framework.
The chart: Cognition’s FrontierCode Diamond benchmark tests whether AI agents can produce production-quality, mergeable code as well as code that passes tests. On the 50 hardest tasks shown in the seeded chart, Claude Opus 4.8 leads at 13.4%, more than 2x GPT-5.5 at 6.3%, while most frontier models stay below 5%.
The lesson: coding agents are moving into reviewer-grade territory, where maintainability and mergeability count more than isolated test passes.
The caveat: 13.4% is still low. The benchmark maps where current agents still fail inside real codebases, even when they look strong in isolated coding tests.
Figure’s Humanoids Pull a Real Warehouse Shift
⚡ Bottom line: Figure ran its F.03 humanoids on a live package-sorting line for days without stopping, fully autonomous on its Helix model.
💡 Why it matters: Short demo clips prove little; an unbroken multi-day shift is the real test of whether humanoids can do useful work.
🔎 What it means: The bottleneck is shifting from one-off dexterity to uptime, reliability, and the manufacturing scale to build fleets.
Figure spent the past month answering the question viral robot clips never settle: can a humanoid hold up across a full shift? The company put its F.03 robots on a conveyor line and livestreamed them sorting packages without stopping, identifying parcels, picking them up, rotating them to scan barcodes, and placing them back, all driven by its Helix model rather than by human teleoperation.
The point of the run was endurance, not party tricks. By day nine of the stream, Figure said the robots had operated for 191 consecutive hours and sorted 238,000 packages at close to human speed, with the system automatically resetting or swapping in another robot whenever one went out of distribution. The broadcast drew millions of views and plenty of skepticism about how staged it was, which is exactly the scrutiny a long public test invites.
Behind the demo sits a manufacturing story. Figure says its BotQ facility has gone from building one Figure 03 per day to one per hour, a 24x jump in under 120 days, and has delivered more than 350 third-generation robots. That pairing is the real signal: the company wants to show both that a single robot can survive repetitive real work and that it can build enough of them to matter.
The IT strategy every team needs for 2026
2026 will redefine IT as a strategic driver of global growth. Automation, AI-driven support, unified platforms, and zero-trust security are becoming standard, especially for distributed teams. This toolkit helps IT and HR leaders assess readiness, define goals, and build a scalable, audit-ready IT strategy for the year ahead. Learn what’s changing and how to prepare.