😺 Apple finally rebuilt Siri

PLUS: Apple turned Siri into an OS layer. Now it has to work.

Author

Grant Harvey
June 09, 2026

OpenAI did the most OpenAI possible sequence yesterday: it filed a confidential S-1 with the SEC, then announced it because it expected the filing to leak. A few hours later, Sam Altman and Jakub Pachocki published a plan for OpenAI’s “third phase,” where the economy starts reshaping around AI, and the company will try to build steerable AI researchers by March 2028.

Gary Marcus read all of this a bit differently. He highlighted a Wharton / NBER paper arguing the capex math only works if AI-sector productivity rises about 2.7x fast enough to justify all the money pouring in. Meanwhile, the market is treating SpaceX, OpenAI, and Anthropic IPOs like the next Google or Amazon moment, even though the valuation math gets WILD, fast.

Dwarkesh Patel offers a much-needed grounding to all this: it seems data and distillation (where a larger model is “distilled” down to a smaller one that’s more efficient to run) may keep AIcapabilities climbing faster than skeptics expect. 

So today’s AI economy has two clocks running at once: the capability clock and the capital clock. The first one’s like a count-down to take-off, and the second’s like an hour-glass where the sand is slowly draining out. 

And speaking of sand running out situations, this new work from MIT on the top AI risks and who is responsible for them is worth a look… read more below!

Here’s what happened in AI today:

  • 🙀 Apple turned Siri into an AI operating layer.

  • 📰 Anthropic showed Mythos can exploit flaws in hours.

  • 📰 NVIDIA made South Korea an AI factory showcase.

  • 🍪 Kimi Work launched 300 desktop agents for knowledge work.

  • 🔧 Greg Isenberg mapped AI-native company systems.

Hey: Want to reach 700,000+ AI-hungry readers? Advertise with us! 

P.S: Love robots? We’re starting a new robotics newsletter! Sign up early here.

🙀 Apple used WWDC26 to turn Siri into an AI operating layer for all your apps… and we’re actually bullish

Apple’s AI problem has been painfully practical for the past three years: the company owns the phone, laptops, watches, headsets, apps, and work stations people already use all day. 

All they need is an intelligence layer on top of it. Siri just wasn’t it. 

Well, at WWDC26, Apple made its reset pitch: AI should live inside the OS, where it can understand your screen, your files, your apps, and your personal context. And they’re finally ready to deliver. 

Here’s what happened:

  • Apple introduced Siri AI, a rebuilt assistant that can hold richer conversations, search messages, emails, photos, and files, and take action inside apps.

  • Apple said its next-gen Apple Foundation Models were custom-built with Google using Gemini technologies, then adapted for on-device use and Private Cloud Compute.

  • Apple added AI features across Safari, Passwords, Messages, Mail, Calendar, Phone, Home, Shortcuts, Image Playground, and Photos. The Newsroom release also confirms SynthID watermarks for AI-edited and AI-generated images.

  • The developer story got deeper in the Platforms State of the Union: Foundation Models now support images, server models like Claude and Gemini, Dynamic Profiles, Evaluations, and no-cost Private Cloud Compute access for smaller developers (more about that here).

How to try it:

Why this matters: Apple is trying to make AI useful by putting it where the friction lives: the camera, keyboard, browser, calendar, photos, files, calls, apps, and screen.

That could be more practical than a standalone chatbot, because Siri can act on your device instead of only answering inside a chat. Ask about a schedule, and it can add events. Ask about files, and it can compare them. Ask about photos, and it can find, edit, and share them. Basically, the best AI application is the AI app that can use all your other apps for you. 

Our take: The pitch is strong. The execution has to be boringly reliable. The hardest part is developer adoption: devs are only now adopting Ai en-masse, and they’re getting addicted to their tools of choice. Siri AI only feels smart outside Apple’s own apps if developers adopt it and wire their apps into the new plumbing.

If it works, the future of AI on Apple devices looks less like “open chatbot” and more like “ask your computer to handle the annoying part.”

Apple’s AI future now depends on a very Apple question: can invisible infrastructure make the computer FINALLY feel easier to use? If they can do this, then we’re finally in the “value-acrues to the app” era of the AI transition. 

How much product risk is hiding in your next release?

Coding agents help you ship faster – but every release creates more flows, edge cases, and bugs that manual testing can't always cover.

Issues take root in the places you didn't think of, didn't have time for, or assumed were working – affecting users, conversions, and release confidence.

For AI-assisted development, you need AI testing. QA.tech's agents explore your product like real users and validate critical journeys across your product, store, or latest release.

Run a free Product Quality Audit and uncover the issues that could impact customers before your next release.

Today’s skill comes from Boris Cherny’s advice for running Claude Opus autonomously for hours or days. The useful idea is to treat autonomy like a system, not a wish: give Claude permission to keep moving, give it a goal loop, then make it verify the finished work.

Set Claude to auto mode so it does not ask for approval on every safe project action. Run it in the cloud so the job keeps going after you close your laptop. Use /goal or /loop, which are steering commands that nudge the agent to continue until the task is done. For bigger work, use dynamic workflows so Claude can coordinate many sub-agents. Then add end-to-end verification: Claude in Chrome for web work, an iOS or Android simulator MCP for mobile (MCP means a tool connection the model can use), or the full running server for backend work. 

Run this as a long-horizon task.


Use auto-approved permissions only for safe project actions.
Use /goal or /loop to keep working until the outcome is complete.
If the task is too large, create a dynamic workflow and split it into sub-agents.
Do not report "done" until you self-verify end to end:
- Web: test in the browser.
- Mobile: test in an iOS or Android simulator MCP.
- Backend: start the full service and run the relevant checks.


At the end, give me:
1. What changed
2. How you verified it
3. What risks remain

Total AI beginner? Start here (goes with this video).

Have a specific skill you want to learn? Request it here. 

📰 Around the Horn

  • Anthropic’s Mythos can exploit newly disclosed software flaws in hours, while its N-day research measured how models accelerate known-vulnerability attacks.

  • NVIDIA and LG anchored South Korea’s AI factory push, with SK hynix memory work and NAVER scaling toward gigawatt infrastructure.

  • Microsoft shut down more than 70 GitHub repos after the Miasma worm targeted Claude and Gemini coding-agent users.

  • Cognition launched FrontierCode, a coding benchmark that asks whether real maintainers would merge AI-written code.

  • Pentagon accused Alibaba, Baidu, BYD, and Unitree of supporting China’s military, escalating AI and robotics security tensions.

  • MIT FutureTech ranked AI risks with 272 experts, then mapped which institutions should own mitigation responsibilities in its responsibility visualization.

NEW FROM THE NEURON:

Want a computer you can actually train your own AI on? Dell Pro Max with GB10 brings NVIDIA Grace Blackwell architecture, 128GB memory, and NVIDIA DGX OS 7 to founders, builders, and AI enthusiasts testing agents, demos, and model workflows locally.

🔧 Tuesday Tool Tip: Make your company readable to agents

Greg Isenberg and Theo Tabah’s AI-native masterclass had one useful rule: using ChatGPT is step one. Building an AI-native company means giving agents the goals, tools, skills, and context they need to work without constant babysitting.

Try the lightweight version this week:

  • Pick one recurring output, like a weekly report, proposal, product brief, or research memo.

  • Create a folder with great examples, customer notes, quality standards, and “do / don’t” rules.

  • Give the agent one clear goal, the tools it can use, and the test for what “good” means.

  • After each run, save the human corrections back into the folder so the system improves.

  • BONUS: do this for each recurring task, but save them as “skills” instead of a project folder (ask the Ai to do it for you), and you can use the Project folder as your work context hub where you can call skills at any time like magic spells. 

The goal is speed with signal: agents move faster, humans judge quality, and the context layer gets smarter every week.

A Cat’s Commentary

“not super-boring” is my new favorite compliment

That’s all for now.

What'd you think of today's email?

Login or Subscribe to participate in polls.

P.S: Before you go… have you subscribed to our YouTube Channel? If not, can you?

P.P.S: Love the newsletter, but only want to get it once per week? Don’t unsubscribe—update your preferences here.