Windows Opens Door To Agents; Scorsese In On AI; Alphabet(s) Big

Today's AI Outlook: 🌤️

Now Windows Wants To Be The Boss Of Your Agents

Microsoft used Build 2026 to make a loud point: its AI future is no longer just a very expensive group project with OpenAI. The company rolled out seven in-house MAI models, a proactive Teams agent called Microsoft Scout, agent-focused device concepts and new AI hardware, all aimed at turning Windows and Microsoft 365 into the command center for autonomous software workers.

Why It Matters

Microsoft is trying to own the agent layer before anyone else names it, brands it or charges a monthly seat license for it.

The Deets

• Microsoft released seven MAI models across reasoning, coding, image, voice and transcription.
• Microsoft Scout, built on OpenClaw, can schedule meetings, prep materials and take proactive action inside Teams.
• Project Solara previews an agent-first platform for new devices, including a badge and desk companion.
• Microsoft also introduced the Surface RTX Spark Dev Box, a mini-PC built for AI workloads.
• Nvidia’s RTX Spark push brings powerful local AI compute to PCs, with Lenovo, HP, Microsoft and Adobe all leaning in.

Key Takeaway

The AI desktop is becoming the new data center, only with more Teams notifications.

🧩 Jargon Buster - Agentic AI: AI systems that can take actions across apps and workflows, rather than simply answering prompts.

♟️ Power Plays

Washington Picks The Soft-Touch AI Button

President Donald Trump signed an executive order asking frontier AI labs to voluntarily submit advanced models for a 30-day government security review before release. The final version backed away from a stricter 90-day draft, while also ruling out mandatory licensing or permits for new models.

Why It Matters

The White House wants visibility into frontier AI risks without slowing the U.S. race against China. That makes this less of a hard regulatory wall and more of a velvet-rope checkpoint.

The Deets

• The review applies to “covered frontier models” flagged through a classified process.
• The order asks labs to participate voluntarily before launch.
• It directs the Justice Department to target AI-powered hacking.
• Former AI czar David Sacks reportedly supported the shorter 30-day review after pushing back on the earlier version.

Key Takeaway

AI regulation is entering its “please cooperate, but we are watching” phase.

🧩 Jargon Buster - Frontier Model: A highly advanced AI model near the leading edge of current capability, often powerful enough to raise national security or cybersecurity concerns.

Anthropic Buys The Pipes Everyone Was Using

Anthropic is buying Stainless, the company that helps developers turn APIs into ready-made SDKs and connectors. Stainless has been used by hundreds of companies, including OpenAI, whose Python, Node, Go and Ruby libraries were generated through the platform.

Why It Matters

Models get the headlines, but developer infrastructure is where loyalty gets built. Anthropic proposed MCP as an open connector standard, and now it owns a major toolchain that helps companies build around APIs.

The Deets

• The deal is reportedly above $300M.
• Stainless is expected to shut down its platform on Sept. 1, 2026.
• Rival AI companies will need to maintain their own SDKs or find replacements.
• The move strengthens Anthropic’s position in the developer ecosystem around tools, connectors and model infrastructure.

Key Takeaway

The model war is moving underground, and Anthropic just bought a very useful shovel.

🧩 Jargon Buster - SDK: A software development kit, which gives developers prebuilt tools and libraries to connect apps to a platform or API.

🛠️ Tools & Products

Scorsese Gives AI Storyboards The Auteur Nod

Martin Scorsese has gone public as an adviser to Black Forest Labs, using the company’s FLUX image model to help storyboard a new film. His use case is limited to preproduction, with no AI-generated actors, sets or footage.

Why It Matters

Hollywood remains deeply suspicious of generative AI, but Scorsese’s endorsement gives the technology a more palatable creative lane: helping artists plan, visualize and communicate ideas without replacing performers or production crews.

The Deets

• Scorsese signed on last year as a Black Forest Labs partner and adviser.
• He is using FLUX for storyboarding only.
• The pitch is speed and creative flexibility, not AI-made cinema.
• His support could soften resistance among filmmakers who are open to tools but wary of replacement.

Key Takeaway

AI may have found its least radioactive Hollywood job: sketching the movie before anyone yells “action.”

🧩 Jargon Buster - Storyboarding: The process of visually mapping scenes before filming, usually through sketches or images that show shots, movement and sequence.

💸 Funding & Startups

Alphabet Opens The AI Infrastructure Wallet

Alphabet is selling $80B in stock to help fund its AI infrastructure expansion, with spending plans reportedly reaching up to $190B this year.

Why It Matters

The biggest AI companies are turning infrastructure into a financial endurance sport. Chips, data centers and power are now strategic weapons, and they are very expensive ones.

The Deets

• Alphabet is raising money through a major stock sale.
• The funding is tied to AI infrastructure expansion.
• The company’s expected spending could reach $190B this year.
• The move reflects how capital-intensive the AI race has become.

Key Takeaway

The AI boom has entered the “bring $80B or do not bother” stage.

🧩 Jargon Buster - AI Infrastructure: The hardware, data centers, networking and energy systems needed to train and run advanced AI models.

🧪 Research & Models

Mythos Makes Zero-Days Look Like A Pricey Compute Problem

Anthropic’s unreleased Mythos model is already drawing attention in cybersecurity. Palo Alto Networks, one of the first in cyber to get access, reportedly used it for three weeks and surfaced more than 20 critical low-level vulnerabilities, at a token cost north of $1M.

Why It Matters

AI may radically speed up vulnerability discovery, which is great for defenders and terrifying for everyone else. The catch is that the same capability can help attackers find the same flaws faster.

The Deets

• Mythos reportedly found vulnerabilities at about five times the rate of traditional tools.
• Anthropic has priced Mythos at six times Opus 4.8.
• Security leaders are still willing to pay because the payoff could be massive.
• The unanswered question is the false-positive rate, which will decide whether this scales cleanly.

Key Takeaway

Cybersecurity is becoming an AI arms race where the entry fee is a giant token bill.

🧩 Jargon Buster - Zero-Day: A software flaw unknown to the vendor, which means attackers can exploit it before a patch exists.

Sutton Points AI Back Toward The Real World

Richard Sutton and Banafsheh Rafiee proposed enactive AI, a path where intelligence emerges from agents acting in the world and learning from consequences. The idea emphasizes experience, action, autonomy and embodiment instead of static training data.

Why It Matters

The paper challenges today’s dominant AI recipe, but it also highlights a brutal engineering problem: real-world learning is slow, messy and hard to scale.

The Deets

• The paper argues intelligence should come from real experience, not only labeled data.
• It treats perception and action as one loop.
• It emphasizes autonomy, where systems set their own success criteria.
• It views reinforcement learning as the closest existing branch, but still incomplete.

Key Takeaway

Enactive AI offers a compelling map of intelligence, though the road crew has not arrived.

🧩 Jargon Buster - Embodiment: The idea that intelligence develops through interaction with a body or environment, rather than only through text or data.

⚡ Quick Hits

• OpenAI Codex added Sites for hosted websites and apps, plus role-specific plugins for broader white-collar work.
• Cognition rebranded Windsurf IDE as Devin Desktop, creating one surface for local and cloud agents.
• Anthropic expanded Project Glasswing, giving 150 organizations across 15 countries access to Claude Mythos Preview.
• Intel is backing a 36,864-core rack design for agentic AI workloads.
• North Korea is targeting open-source maintainers, raising supply-chain security risks for AI-era software.
• Uber capped employee AI spending after burning through its budget in four months.
• Google rolled out fake-call detection to protect users from AI deepfake impersonation scams.
• Amazon Ring faces a class-action fight over facial recognition.

🧰 Tools Of The Day

• MAI: Microsoft’s new in-house model family covers reasoning, coding, image, voice and transcription.
• Claude Design: A slide deck workflow that turns messy reports into strategy decks with charts and recommendations.
• Holo3.1: H Company’s upgraded computer-use model runs fully locally.
• Hermes Desktop: Nous Research’s agent runs as a native desktop app.
• Codex: OpenAI’s agentic coding tool gained Sites and role-specific plugins.
• MyClaw: A cloud service that turns OpenClaw into an always-on operator connected to apps like Gmail, Slack, Notion, GitHub and WhatsApp.

Today’s Sources: The Internet, The Rundown AI, AI Secret