😸 Microsoft's AI "operating system"

Welcome, humans.

Quick house-keeping note: Thanks for your patience with today’s later send; we’re in person at Microsoft Ignite, it was a HUGE news day, and also there was a global Cloudflare outage, so lots to navigate!

The hotly anticipated new model from Google, Gemini 3 just launched. This newsletter was written yesterday, before Gemini 3 was official, but don’t worry, we’ll have more content on that coming out later today with a quick TL;DR down below.

So Meta just announced it'll start grading employees on their “AI-driven impact“ starting in 2026. Translation: use AI to crush your job, or expect it to show up in your performance review.

We've been watching this trend build across Big Tech. Microsoft told managers AI is “no longer optional.“ Google's CEO said employees need to use it for Google to lead the AI race. Now Meta's making it official—your bonus could depend on how well you wield AI tools.

Fun fact: Meta already launched an internal game called “Level Up“ to incentivize AI adoption. Gamifying performance reviews? What could possibly go wrong! (Actually, this might be brilliant. Or terrifying. We'll find out in 2026...)

Here’s what happened in AI today:

1. We cover everything Microsoft announced at Ignite 2025.
2. Jeff Bezos launched Project Prometheus with $6.2B.
3. Memory chip shortages loom for phones and cars
4. Gemini 3.0 has arrived (more to come on this later!).

Microsoft Just Built the “Operating System“ for AI Agents

DEEP DIVE: Everything Microsoft Announced at Microsoft Ignite

Companies are drowning in AI agents. Sales agents. Customer service agents. Data analysis agents. HR agents. Some built by IT, most created by employees who found a cool tool online (maybe even something we showed y’all!). Nobody knows what's running, what data agents can access, or which ones might be security risks.

Microsoft just solved this at Ignite 2025 with Agent 365, a control plane that discovers, manages, and secures every AI agent in your organization. Think of it as the IT admin dashboard for your robot workforce.

Here's what makes this a big deal:

• Agent 365 creates a complete registry of every agent in your org including “shadow agents“ that employees created on their own. No more surprises when Finance discovers Marketing's been running an unapproved agent with access to customer data for three months.
• It applies risk-based access controls to agents just like employees. IT can limit agents to only the resources they need, block risky agents before they cause problems, and set conditional access policies based on agent behavior.
• The system offers real-time monitoring showing connections between agents, people, and data. IT teams can finally see what agents are actually doing and measure their impact on the organization.

Oh, also? Security is built in: Agent 365 integrates Microsoft Defender for threat protection, Microsoft Entra for identity management, and Microsoft Purview for data governance. Agents get the same enterprise-grade security as human employees.

Microsoft also announced concrete examples of what these managed agents look like:

Sales Development Agent (available now via Frontier program): A fully autonomous sales agent that researches prospects, crafts personalized outreach, and follows up 24/7. When leads are qualified, it hands them off to human sellers. Basically, an SDR that never sleeps.

Word, Excel, and PowerPoint Agents in Copilot Chat: Create high-quality Office content by asking targeted questions to understand your goals. They handle research, formatting, and layout design—no more staring at blank documents.

12 new Security Copilot agents embedded in existing security tools to automate alert triage, identity risk management, endpoint compliance, and data remediation.

Oh yeah, and guess who made it official? Anthropic is now officially part of the Azure ecosystem: Microsoft added Anthropic’s Claude 4.5 Sonnet, Haiku, and Opus to its library. It is now the only cloud provider hosting both OpenAI and Anthropic.

Flex.

Why this matters: A new IDC study commissioned by Microsoft surveyed 4K+ business leaders and found something striking. While 68% of companies use AI today, there's a massive gap forming.

The top 22% (called “Frontier firms“) are achieving returns three times higher than slow adopters. What separates them? They use AI across an average of seven business functions (not just one or two), they're building custom AI solutions (58% today, jumping to 77% in 24 months), and they're aggressively adopting agentic AI.“

IDC estimates the number of companies using agentic AI will triple in the next two years (the number quoted at Ignite was 1.3 billion agents deployed by 2028. So the window to become a Frontier firm is closing.

Agent 365 gives organizations the infrastructure to scale agents without losing control and solving the exact concerns (security, governance, visibility) that have kept IT departments from embracing autonomous AI.

It's available now through Microsoft's Frontier program.

Check out the rest of our coverage on Microsoft Ignite here.

FROM OUR PARTNERS

The Battle Against Bots: How to Protect Your AI App

Modern bots are smarter than ever. They execute JavaScript, store cookies, rotate IPs, and even solve CAPTCHAs with AI. As attacks grow more sophisticated, traditional detection isn’t enough.

Enter WorkOS Radar: your all-in-one bot defense solution. With a single API, you can instantly secure your signup flow by blocking bots, preventing brute force attacks, and catching free trial abuse.

Stop threats before they start. Keep real users flowing through.

Protect your app today →

Prompt Tip of the Day

Tired of AI giving you mediocre outputs because it picked the easiest path? This “task decomposer“ prompt from Excellent AI Prompts forces the model to propose multiple solution methods, compare them side-by-side, and justify its choice before generating anything.

Here's how it works: Instead of letting the AI jump straight to execution, you make it act as a “systems designer“ that must first propose two different approaches. For each method, it lists 5 specific steps, estimates time required, and identifies the biggest risk. Then it creates a comparison table showing strength, risk, speed, and quality for each approach. Only after choosing one method with a clear rationale does it execute.

Our favorite insight: Most bad AI outputs happen because the model confidently executes the first weak approach that came to mind. By forcing method selection before execution, you get intentional workflow design instead of whatever random path the AI stumbled down first.

Treats to Try

*Asterisk = from our partners (only the first one!). Advertise to 600K daily readers here!

1. *The Dell Pro Max with GB10 Grace Blackwell Superchip packs next-gen GPU architecture, 128GB RAM, and 4TB storage into a desktop that handles model training and data science workloads other machines can't touch. While everyone else waits for Blackwell, you're already building with it.
2. Explore Dell Pro Max →
3. Grok 4.1 writes X posts and essays with you (w/ a 600-point creative writing improvement), understands emotional context better, and makes 3x fewer errors while ranking #1 in quality—try it here.
4. Aurasell replaces your entire sales stack of 15+ tools with one unified platform that handles everything from finding leads to closing deals.
5. Cyera scans your entire data environment in minutes to discover and protect your sensitive data across all your systems, shows you exactly where sensitive info lives and who has access to it.
6. Sakana AI creates affordable AI models optimized for Japanese language and culture that work efficiently with small datasets for enterprises in finance, manufacturing, and government (raised $135M).
7. Turing develops AI-powered autonomous driving systems adaptable to various vehicle types, working with Nvidia to launch fully self-driving commercial vehicles by 2029 (raised $99M)
8. Reindeer works inside your IDE to understand your database schema and generate production-ready SQL queries in seconds, with real-time autocomplete and automatic error fixes as you type.
9. BeFreed turns learning into personalized, conversation-like experiences using AI to make education feel natural and joyful rather than tedious and rigid.

Around the Horn

Well, this is cheery: Anthropic's CEO just told 60 Minutes that AI might eliminate half of all entry-level white-collar jobs within five years, potentially doubling unemployment. Sweet dreams, everyone!

• Jeff Bezos launched Project Prometheus with $6.2B in funding, taking his first operational role since leaving Amazon to build AI for engineering and manufacturing in computers, aerospace, and automobiles.
• Bone AI builds autonomous drones, ground vehicles, and marine vessels for defense missions like logistics support, wildfire detection, and anti-drone defense (raised $12M).
• Chipmakers warned of memory chip shortages for consumer electronics and cars as manufacturers prioritize AI server demand, with Samsung raising prices up to 60%.
• Three AI-generated songs topped Spotify and Billboard charts as 50,000 AI tracks upload daily to streaming platforms.
• Nitro's new survey reveals your C-suite executives are secretly using ChatGPT on personal accounts; 68% admit bypassing the enterprise tools they spent millions on because approved platforms can't match consumer tools for speed and simplicity.

FROM OUR PARTNERS

Your AI Is only as smart as your Data. Get ready with CData Sync.

Start Your Free Trial Today!

ThursdAI = now TuesdAI?

It used to be popular to call Thursdays “ThursdAI” because companies would often ship their big releases on Thursday. TBH, these days everybody seems to ship everything as soon as it’s ready to go out the door. Tuesday is the big day this week with Microsoft’s Agent Foundry and Google’s release of Gemini 3.0, which we’ll dive deeper into later today. For now, here’s a quick TL;DR of what you need to know to get started:

1. Google launched Gemini 3, their most intelligent model featuring state-of-the-art reasoning that tops LMArena with 1501 Elo and introduces Gemini 3 Deep Think for enhanced reasoning (achieving 45.1% on ARC-AGI-2)…
2. Which powers the redesigned Gemini app with generative interfaces and a new Gemini Agent for multi-step tasks…
3. Brings advanced search capabilities to AI Mode with improved query fan-out and on-the-fly generative UI including interactive tools and simulations…
4. Is now available for enterprise through Vertex AI and Gemini Enterprise with major customers reporting 35-50% performance improvements…
5. Offers developers unprecedented agentic coding capabilities (scoring 1487 Elo on WebDev Arena and 76.2% on SWE-bench Verified, which seems to be the only non-frontier benchmark for Gemini) through Google AI Studio, Vertex AI, and the new Google Antigravity agentic development platform…
6. …And unlocks generative UI research that dynamically creates immersive visual experiences and interactive interfaces completely customized for any prompt.

RE: SWE-Bench, which is a coding benchmark, Google DeepMind researcher Vedant Misra put it best:

A Cat’s Commentary