AI news June 7, 2026: Anthropic wants to freeze AI, Microsoft launches Scout, Supabase hits $10 billion
Alexandre
··
Reading time: 10 min
Click to enlarge
Last week, Anthropic filed its IPO for a $965 billion valuation, Claude Opus 4.8 dropped with 1,000 parallel sub-agents, and Gemini Spark went 24/7. We thought June was starting strong. Turns out the sequel is even crazier.
Supabase doubles its valuation to $10.5 billion in 8 months. Microsoft launches Scout, an "AI coworker" that works while you sleep, built on OpenClaw. Apple is preparing an AI agent App Store for Monday. And the cherry on top: Anthropic (the same company that just filed the biggest IPO in history) publishes a paper calling for a global freeze on AI development because Claude is approaching recursive self-improvement.
This isn't just a week of news. It's the week AI agents went from prototype to product, the cash kept pouring in, and the question "are we going too fast?" officially went mainstream. As a solo dev building Waku with Claude Code, half of these announcements directly impact my daily workflow.
Let's break it down.
Supabase hits $10.5 billion: vibe coding is blowing up the charts
Supabase, the open-source backend platform positioned as the Firebase alternative, just raised $500 million in a Series F at a post-money valuation of $10.5 billion according to Axios. In 8 months, the startup doubled its valuation, and CNBC directly attributes this explosion to the "vibe coding" phenomenon, the practice of building apps by steering an LLM rather than typing every line (CNBC).
The round was led by GIC, Singapore's sovereign wealth fund. In October 2025, Supabase was valued at $5 billion. 8 months and a viral phenomenon later, it's worth double.
Vibe coding has become the default mode for a new generation of developers. And Supabase is their go-to backend: managed Postgres, auth, storage, real-time, everything you need to launch a project in 20 minutes with an AI agent. TechCrunch notes that the massive adoption comes from developers building AI apps who need a backend that works immediately, zero configuration.
What I think
What strikes me isn't the valuation itself, it's the speed. $5 billion to $10 billion in 8 months, for managed Postgres. That's the signal that vibe coding is no longer a Twitter meme, it's an economic movement.
For us indie devs, it's good news. The bigger tools like Supabase get, the more they invest in DX, the more our productivity goes up. When I'm coding Waku, every minute saved on backend setup is a minute more on the feature that matters.
The flip side: when a company raises $500 million, it has to justify that valuation. That means enterprise pricing, premium features, and maybe someday the end of the generous free tier. Open-source Postgres is still there, but the "all-inclusive free" has an expiration date.
While Supabase counts its billions, Microsoft is building something far more ambitious.
Click to enlarge
Microsoft Scout: the AI coworker that never logs off
Microsoft unveiled Scout at Build on June 2: an "always-on" personal AI assistant built on the open-source OpenClaw framework and powered by MAI-Thinking-1, the first reasoning model developed internally by Microsoft. 35 billion active parameters, trained exclusively on commercially licensed data (Axios).
Scout integrates across the entire Microsoft 365 ecosystem: Outlook, Teams, OneDrive, Copilot. It reads your emails, manages your calendar, drafts your replies, reschedules your meetings. According to Omar Shahine, corporate VP at Microsoft, Scout "continuously learns a user's work habits, retains memories and skills, and acts on the user's behalf even when the user is offline" (WIRED).
The project's internal name? "ClawPilot," under "Project Lobster" (404 Media). And it's 404 Media that broke the most embarrassing scoop: internal documents reveal that Microsoft wants to "make people addicted" to Scout. The exact word. Not "engaged." Not "loyal." Addicted. The product is being deployed through the Frontier program and requires a GitHub Copilot subscription.
My take
OpenClaw was the early adopters' toy in January. 5 months later, it's the engine of an enterprise Microsoft product. The speed of absorption is staggering, and it's exactly the pattern I described in my analysis of AI agent architecture: open frameworks always end up absorbed by the giants.
What bothers me is the word "addicted" in the internal docs. An assistant that works while you sleep, remembers everything, acts on your behalf, that's either the best productivity tool ever invented or the worst nightmare in terms of digital dependency. And Microsoft has the involuntary honesty of saying it's aiming for the latter. An AI agent should amplify your will, not replace it.
The real question: will Scout be good enough that we accept the deal? If yes, the personal AI assistant market just found its standard. If not, it'll be Clippy 2.0 with a massive marketing budget.
And while Microsoft builds the AI assistant for the office, Apple is preparing something potentially more important for developers.
Click to enlarge
Apple WWDC Monday: Siri powered by Gemini and an AI agent App Store
Apple is about to unveil Monday at WWDC 2026 a major overhaul of Siri, powered by Google's Gemini model, with agentic capabilities (the ability to delegate multi-step tasks like booking a restaurant or editing a document). But the real bombshell, reported by The Information via TechCrunch, is the planned introduction of a dedicated App Store for third-party AI agents.
According to MacRumors, Siri 2026 will be a standalone app with a chat-style conversational interface, conversation history, and automatic deletion timers. The new "Visual Intelligence" in the Camera app will use image search to identify objects in real time.
On the developer side, the Shortcuts overhaul changes everything: you'll be able to ask Siri in natural language to create multi-step automations. "Send me a summary of my unread emails every morning at 8am," Siri generates the Shortcut. Writing Tools gain a "Help Me Write" mode and an AI-powered grammar checker across Messages, Mail, and all Apple apps.
But the real paradigm shift is the AI agent App Store. Users will be able to choose among different third-party chatbots to respond to their queries, an unprecedented opening of the Apple ecosystem to LLMs. If Apple creates an agent marketplace, distribution changes for everyone.
My take
OK, I'm both excited and puzzled. Excited because I've been developing iOS apps for two years, and an AI agent App Store is potentially the biggest distribution shift since the original App Store in 2008.
Puzzled because Siri powered by Gemini means Apple publicly admitting its in-house AI isn't good enough. Apple Intelligence never convinced anyone. Now they're going to Google for the language model. It's pragmatic, but it's also an admission of failure.
For devs like me who build iOS apps, the agent App Store opens a completely new distribution channel. Imagine Waku, my visual PKM, accessible directly from Siri. "Hey Siri, save this info in Waku." If Apple does this right, it's a revolution for indie iOS devs.
The big unknown: access conditions. Will Apple require a review process for AI agents like they do for apps? If so, the delays and arbitrary rejections will start all over again. Fingers crossed.
But while everyone's preparing their AI products, the company building them just dropped a philosophical bomb.
Anthropic calls for a global freeze on AI development
On June 4, Anthropic (the same company that filed its IPO on June 1 for what could be the biggest public offering in history) published a warning: its most advanced models are approaching recursive self-improvement capability, a threshold where AI can improve itself without human intervention. Anthropic is calling on the world's leading labs to slow down or temporarily freeze development of the most powerful models (SiliconAngle).
Recursive self-improvement is the scenario where an AI model can modify its own code, its own weights, or create better successors without a human overseeing the process. This isn't science fiction, it's the theoretical concept most dreaded by alignment researchers for the past decade.
PYMNTS reports that Anthropic described the risk as "massive societal disruption." Fortune details that the company's researchers documented "dramatic increases in productivity": engineers are producing significantly more code with the latest models. But the shift to fully autonomous AI could cause "humans to lose control" if no verification mechanisms are put in place.
Hindustan Times reports that Anthropic stated a global slowdown would be "likely a good thing." The company plans to organize conversations in the coming months with policymakers, researchers, and other stakeholders to design verification mechanisms before autonomous self-improvement becomes routine.
What I think
Let's pause for a second and measure the absurdity of the situation.
Anthropic files an IPO on Monday and calls for a global freeze on Thursday. That's the most spectacular cognitive dissonance in tech history.
Either Anthropic is sincere and that's terrifying, it would mean the researchers who know these models best think we're approaching a point of no return. Or it's strategic positioning: "we're so far ahead we can afford to call for a pause, and it'll hurt our competitors more than us." Both options are scary.
For me, a solo dev who uses Claude Code every day (I wrote about it in detail here), this is a concrete reality. Opus 4.8 with its 1,000 sub-agents is already a system that does things I don't fully understand. And I'm not talking about a theoretical model, I'm talking about the tool I use to code Waku every day. When the company that makes your primary work tool says "maybe we should slow down," you listen.
The problem is that a global freeze is a utopia. OpenAI won't stop. Google won't stop. And Chinese labs even less so. It's the prisoner's dilemma at planetary scale: whoever stops first loses. Unless everyone stops at the same time. And in human history, that's never happened.
The market has decided that existential risk isn't its problem. For now.
Trump signs a lightweight AI order focused on cybersecurity
On June 2, President Trump signed an executive order titled "Promoting Advanced Artificial Intelligence Innovation and Security," a narrow directive focused on AI cybersecurity after canceling a more ambitious text weeks earlier. The order creates a voluntary federal security testing framework: developers can submit their most powerful models to the government 30 days before release (Politico).
According to CSO Online, the order strengthens criminal penalties for using AI for hacking or other cybercrimes, and creates a federal clearinghouse (a centralized hub for sharing AI cybersecurity intelligence) for critical infrastructure like banks and hospitals (Newsweek).
The key word: voluntary. Nobody is required to submit their model. It's an invitation, not a regulation. When you compare it with Anthropic's call for a global freeze, the gap is striking. On one side, the most advanced lab says "we should stop." On the other, the government says "send us your models if you want, no obligation."
Other news in brief
Alphabet raises $84.75 billion for AI: Google upsized its equity offering to $84.75 billion to fund its AI infrastructure, including $10 billion invested directly by Warren Buffett's Berkshire Hathaway (Reuters). Annual capex budget revised to $180-190 billion for 2026. The number is so large it's hard to visualize, it's more than the GDP of most African countries.
SK Hynix and Micron join the $1 trillion club: both memory chip makers surpassed $1 trillion in market cap, driven by explosive demand for HBM (High Bandwidth Memory, the ultra-fast memory used in AI accelerators) (Reuters). SK Hynix reached $1.34 trillion. AI hardware is no longer a subcontractor, it's an empire.
Ramp valued at $44 billion: the fintech platform raised $750 million at a $44 billion valuation, tripling its worth in a year (TechCrunch). Annualized revenue above $1 billion, positive free cash flow, 70,000 clients including Visa, Uber, and Shopify. AI in fintech pays.
OpenAI expands Codex beyond code: on June 2, OpenAI launched "Sites" (interactive web pages generated by Codex), "Annotations" (context highlighting in documents), and 6 agentic plugins for specific professions: analytics, creative, sales, product design, equity investing, and investment banking (TechCrunch). Codex now exceeds 5 million weekly active users.
Supply chain attack on Codex: a malicious npm package named "codexui-android" was caught stealing Codex authentication tokens from developers (The Hacker News). A brutal reminder that the attack surface grows as fast as adoption.
What all this means for us indie devs
1. AI agents are going from lab to office, and it's irreversible
Microsoft Scout, Apple WWDC, OpenAI Codex with 6 professional plugins: in a single week, three giants showed that AI agents are no longer a concept. It's a product. A product with a price, an "install" button, and a VP who says "we want to make people addicted." The train has left the station.
2. Vibe coding has become a legitimate economic engine
When Supabase doubles its valuation to $10 billion because devs are coding with LLMs, it's no longer a meme, it's a market segment. Tools that simplify the backend for AI developers are capturing a disproportionate share of funding rounds. If you're building tools for devs, the time is now.
3. The dissonance between profit and caution has reached unprecedented levels
Anthropic files an IPO on Monday and calls for a global freeze on Thursday. Google raises $85 billion for AI while knowing its own researchers publish papers on the risks. The market's message is clear: everyone knows this is accelerating dangerously, nobody wants to stop first.
4. Apple's ecosystem is finally opening to agents, and that changes distribution
If Apple launches an AI agent App Store, it's the biggest change in iOS app distribution since in-app purchases. Indie devs already in the ecosystem have a first-mover advantage. Now's the time to think about how your app can become an agent.
Conclusion: the week everyone accelerated (except the one who said stop)
In one week, we watched the market accelerate in every direction. Supabase capturing $500 million. Microsoft launching an agent that never sleeps. Apple opening its ecosystem to chatbots. Trump signing a voluntary decree. And in the middle of all that, Anthropic (the most highly valued company in the group) saying "hey, maybe we should slow down a bit before losing control."
This contrast is the exact image of the AI industry in June 2026. Everyone's running, everyone's investing, everyone's shipping products. And the only company saying "careful" is also the one that just filed the biggest IPO in tech history. The prisoner's dilemma at planetary scale, in real time.
What about you, what are you doing with AI agents in your workflow? Already have a "Scout" or "Codex" running in the background? Hit me up on Twitter/X or in the comments.
Alex
Key takeaways
AI agents are going from lab to office: Microsoft, Apple, and OpenAI are shipping real products this week
Supabase at $10 billion confirms vibe coding is a real economic segment
Anthropic files a record IPO then calls for a global freeze 3 days later, historic dissonance
Apple's AI agent App Store could be the biggest iOS distribution shift since 2008
Trump signs a voluntary AI decree, regulation remains far behind the industry
