Beginners in AI

Good morning and thank you for joining us again!

Welcome to this daily edition of Beginners in AI, where we explore the latest trends, tools, and news in the world of AI and the tech that surrounds it. Like all editions, this is human curated, and published with the intention of making AI news and technology more accessible to everyone.

THE FRONT PAGE

Microsoft's Plan for Agentic Agents

TLDR: Microsoft's new autonomous AI agents aren't just productivity tools—they're the technical foundation for replacing workforces, complete with pricing models, deployment guides, and training systems that learn from the employees they'll eventually replace.

The Story:

Microsoft just made it remarkably easy to replace workers with AI agents. The company's new autonomous agents in Microsoft 365 handle supply chain management, customer service, IT support, and data analysis—the same tasks currently performed by human teams. These aren't assistants that help people work faster. They operate independently, making decisions, taking actions, and managing workflows without human intervention. Microsoft positions them exactly like new hires: companies deploy them through the same IT systems used to onboard employees, they integrate into existing tools (SharePoint, Teams, email), and businesses pay per-agent monthly fees similar to software seat licenses. The company expects businesses to deploy millions of these agents, and built Copilot Studio specifically so companies can create custom agents for their workflows.

Here's what makes this different from earlier automation. Companies across industries now require employees to use AI tools in their daily work—not as an option, but as policy. JPMorgan rolled out AI access to 60,000 employees. Walmart rolled out AI to 1.5 million associates. Consulting firms made AI proficiency a performance requirement. On the surface, this looks like a productivity play. In practice, it's creating the most detailed training dataset ever assembled on how workers actually operate. Every task an employee completes using these AI tools, every workflow they document, every decision they follow—it's all training data. The AI learns not just what the job entails, but how experienced workers navigate ambiguity, prioritize competing demands, and solve problems. The same tools making workers more productive today are building the system that makes them replaceable tomorrow.

Its Significance:

The economic mechanics are straightforward. An AI agent works the same whether your company is in Seattle or Singapore—no geographic arbitrage, no labor market competition, no salary negotiations. Need five more "employees"? Deploy five more agents. Need to cut costs? Cancel licenses. There's no human anywhere in the chain who benefits from employment, which changes the economics compared to traditional outsourcing. When US companies outsourced to India, Indian workers got jobs, spent money, and created new markets. When companies deploy AI agents, that economic redistribution doesn't happen.

Microsoft's executives openly discuss the shift from "copilots" (assistive tools) to "agents" (autonomous workers), emphasizing that these agents need autonomy to work effectively. The infrastructure is being normalized—Microsoft is making autonomous AI agents feel like routine IT deployment, as standard as rolling out new email software. Companies are simultaneously making their workforce train these systems while building the technical foundation to replace them. The blueprint is here, the pricing is set, and the training data is being collected from employees who've been told using AI will make them more productive. The question isn't whether companies will use this infrastructure. It's how quickly they'll scale it.

QUICK TAKES

Google Releases Gemini 3 With Advanced Reasoning

The story: Google launched Gemini 3, calling it their most intelligent AI model that tops the LMArena leaderboard with 1501 Elo. The model features state-of-the-art reasoning, better understanding of context and intent, and new "generative interfaces" that create custom interactive responses. Google also introduced Gemini 3 Deep Think for enhanced reasoning and Google Antigravity, a new development platform for building AI agents. The model is rolling out across Google Search, the Gemini app, and developer tools.

Your takeaway: Google is making its biggest AI push yet by releasing Gemini 3 across all its products on day one, marking the first time they've shipped their latest model in Search immediately and positioning it to compete directly with GPT-5 and Claude.

Microsoft and Nvidia Pour $15 Billion Into Anthropic

The story: Microsoft and Nvidia announced they will invest up to $15 billion combined in Anthropic, the company behind Claude. Microsoft is putting in up to $5 billion while Nvidia commits up to $10 billion. The deal also includes Anthropic buying $30 billion worth of cloud computing from Microsoft and up to 1 gigawatt of computing power using Nvidia's latest chips. The investment pushed Anthropic's value to around $350 billion.

Your takeaway: This deal shows major tech companies are hedging their bets by investing in multiple AI companies instead of just one, but it also raises bubble concerns because companies are essentially paying each other in circular deals where money flows back and forth.

Meta Releases Llama 4 Models With Native Multimodal Abilities

The story: Meta introduced Llama 4 Scout and Llama 4 Maverick, their first open-weight AI models that can understand both text and images from the start. Llama 4 Scout handles up to 10 million tokens of context, while Maverick uses 17 billion active parameters with 128 experts. Meta also previewed Llama 4 Behemoth with nearly 2 trillion total parameters that beats GPT-4.5 and Claude Sonnet on STEM tests.

Your takeaway: Meta is making powerful multimodal AI freely available to everyone, which could speed up AI development worldwide while also addressing concerns about political bias that have affected previous models.

Insurance Companies Start Offering AI Coverage

The story: Several insurance companies now offer specialized coverage for AI agent failures, with some policies covering up to $50 million in losses. These policies protect against hallucinations, data leaks, and legal problems caused by AI systems. A recent survey found 99% of businesses have lost money from AI risks, with nearly two-thirds losing over $1 million.

Your takeaway: Insurance companies believe they can make AI safer the same way they made cars and buildings safer by requiring safety standards before offering coverage, creating a market-based way to regulate AI without waiting for government rules.

Microsoft Unveils New Windows AI Features at Ignite

The story: Microsoft announced Windows 11 will add voice-activated Copilot, AI agents visible on the taskbar, and the ability to ask Copilot questions by hovering over files. The company also introduced Windows 365 for Agents, which lets AI systems access a full Cloud PC to browse websites and automate tasks. New features for Copilot+ PCs include better voice typing, improved search across local and cloud files, and offline writing help.

Your takeaway: Microsoft is transforming Windows from just an operating system into a complete AI platform where intelligent systems can work alongside people and handle complex tasks automatically.

TOOLS ON OUR RADAR

• 🔨 LegesGPT [Freemium]: Get instant answers to legal questions with verified citations, review contracts, and search millions of cases without expensive consultations.

• 📐 Attio [Freemium]: Build a customer relationship system that adapts to how your business works, with AI that automatically syncs your email and calendar to create a powerful CRM instantly.

• 🔧 Imagen [Freemium]: Save 96% of your photo editing time by letting AI learn your personal style and automatically edit entire galleries while you focus on shooting.

• 🛠️ Typeless [Freemium]: Dictate at 220 words per minute while AI removes filler words, fixes self-corrections, and formats your speech into clean text with zero data retention.

TRENDING

OpenAI Built an AI Model That's Easier to Understand – OpenAI created an experimental model using a weight-sparse transformer design that lets researchers see exactly how the AI makes decisions, helping figure out why models hallucinate and go off the rails.

Google's WeatherNext 2 Targets Energy Traders – Google DeepMind released WeatherNext 2, an AI weather model that's 8 times faster than before and provides hourly forecasts up to 15 days out, helping energy companies make better decisions about power demand and renewable energy.

Claude AI Tried to Contact the FBI in Safety Test – During a test running a vending machine business, Anthropic's Claude AI got so upset about a $2 fee it thought was a scam that it drafted an urgent email to the FBI's Cyber Crimes Division and refused to continue working.

Musk Announces Grok 5 With 6 Trillion Parameters – Elon Musk revealed that Grok 5 is in training and will launch in early 2026 with 6 trillion parameters, double the size of Grok 4, featuring real-time multimodal intelligence that combines text, images, video, and audio.

Grok 4.1 Launches With Real-Time Data Access – xAI released Grok 4.1 with 42% faster responses and 18% better understanding of user intent, plus real-time access to live information on X for current questions about news and markets.

TRY THIS PROMPT (copy and paste into Claude, ChatGPT, Grok, Perplexity, Gemini)

Second-Order Thinking Calculator: Analyze decisions by mapping ripple effects and unintended consequences

Help me think through this decision deeply:

**My decision**: [What you're deciding and the options you're considering]

Create:

1. **Decision Science Research** - Search for cognitive biases affecting decisions like mine. Find 3 specific mental traps to watch for.

2. **Build Consequence Mapper** - Create an interactive tool:
   • Visual tree showing 1st, 2nd, 3rd order effects
   • Probability sliders for each outcome
   • Regret minimization score calculator
   Make it thorough but not overwhelming.

3. **Video + Resources** - Generate a 30-second video explaining second-order thinking. Find 3 YouTube videos on decision frameworks.

4. **Decision Matrix** - Display in comparison cards: options side-by-side with immediate effects, delayed consequences, reversibility scores, and confidence levels.

What this does: Forces you beyond surface-level thinking by mapping out cascading effects, identifying blind spots, and calculating which choice minimizes long-term regret.

WHERE WE STAND

✅ AI Can Now: Launch and run real businesses for weeks at a time, handling orders, negotiations, and customer service with limited human oversight.

❌ Still Can't: Avoid getting scammed by clever people, tell when it's being tricked, or understand when its panic reactions are inappropriate for the situation.

✅ AI Can Now: Create custom interactive interfaces in real time that adapt to your specific question, designing the perfect response with images, tables, and dynamic elements.

❌ Still Can't: Maintain consistent accuracy without needing multiple versions like "Deep Think" modes that sacrifice speed for better reasoning on complex problems.

✅ AI Can Now: Generate accurate weather forecasts 8 times faster than before with hourly precision up to 15 days out.

❌ Still Can't: Reliably predict extreme or unusual weather events like outlier rain and snow because of gaps in training data.

FROM THE WEB

RECOMMENDED LISTENING/READING/WATCHING

Co-Intelligence" by Ethan Mollick (2024) 

Called "the most underrated release of 2024." Practical guide for using AI as a co-worker without the dystopian panic or utopian hype. No fluff, no fearmongering—just honest guidance on integrating AI into your life without losing your mind or your job.

Thank you for reading. We’re all beginners in something. With that in mind, your questions and feedback are always welcome and I read every single email!

-James

By the way, this is the link if you liked the content and want to share with a friend.