Edition #2. This week again some new developments from the AI world.
In this edition: - The AI arms race: three model updates and a mega-funding in one week - Anthropic raises $30 billion, valuation $380 billion - Claude Cowork now also available on Windows - The SaaS-pocalypse: why software stocks dropped - What Claude Code, Cowork and AI agents can concretely do for you
Three releases, a mega-funding, and a stock crash. One week.
OpenAI launched GPT-5.3-Codex-Spark. Their fastest coding model ever, built in partnership with chipmaker Cerebras. More than 1000 tokens per second. Real-time coding. Not fast for AI. Just fast. The company behind it all? Currently valued at $500 billion, the most valuable AI company in the world.
Anthropic (makers of Claude) raised $30 billion this week in their Series G, led by Singapore's sovereign wealth fund GIC and investor Coatue. Valuation: $380 billion. That makes Anthropic, together with OpenAI and SpaceX, the trio of the most valuable startups in the world. And all three are being watched this year as possible IPO candidates.
Part of the funding comes from the previously announced $15 billion from Nvidia and Microsoft, tied to a deal where Anthropic takes about $30 billion in compute from Microsoft. The company is also heavily backed by Amazon and Google.
And the revenue at Anthropic? In January 2023: $0. Today: $14 billion in annual revenue. 10x growth every year, three years in a row.
Claude Code now has a run-rate revenue of more than $2.5 billion. The number of weekly active users has doubled since January 1. Doubled. In six weeks.
Claude now also sits directly in Microsoft Word, Excel and PowerPoint via Cowork plugins. Building financial models, analyzing contracts, making presentations. Just from your toolbar. After software developers, just about everyone gets touched by the AI efficiency boom now happening.
MiniMax from China comes out with M2.5
Just as smart as the best models from Anthropic and OpenAI, but 10x cheaper. $1 per hour running non-stop. An AI agent that watches your code 24/7, fixes bugs and opens PRs for $720 per month. That's less than the day rate of a junior developer.
The company recently went public and is going to open-source the weights, so anyone can build with it. The message is clear: top-tier AI is becoming a commodity. And that pushes the price down for everyone.
Google updates Gemini 3 Deep Think
Google is updating their heaviest reasoning model. The model reaches gold medal level at the International Math, Physics and Chemistry Olympiads. A mathematician at Rutgers University used it to find an error in a scientific paper that had passed human peer review. Deep Think is now also available via the Gemini API, so developers can build with it.
What can you actually do with it?
The impact of the model updates is actually very practical. These are tools you can use today.
Claude Code is an AI developer in your terminal. You describe in plain English what you want. An app, a website, an automation. And it builds it. Not a sketch you have to finish yourself. The whole thing.
It tests itself. It clicks through the buttons. It uses the app like a human would. Only when it's satisfied does it come back to you: it's ready to test.
The next step is called agentic engineering: you don't give the AI an assignment, but a goal. The AI plans the steps itself, executes them, tests the result, and iterates until it works. You're the architect, the AI is the entire build team.
Claude Cowork is the same idea, but for everyone allergic to the word code. It works from your desktop, now also runs on Windows, and sits directly inside Microsoft Office. Receipts become an Excel report, ten documents become a summary, and a presentation gets made by asking. No terminal. Just drag and ask.
And now it sits inside Microsoft Office too. Build a financial model in Excel? Ask. Analyze a contract in Word? Ask. Build a presentation in PowerPoint? Ask. It sits in your toolbar as if it had always been there.
For office workers: search through months of emails. Analyze entire annual reports. Summarize a quarter's worth of customer feedback. Things that normally take days, you now do in minutes.
And the market feels it. Since the launch of Cowork plugins, SaaS stocks dropped hundreds of billions in market value. If AI can check contracts, analyze data and do compliance, what are you still paying tens of thousands of euros per year in software licenses for? A SaaS-pocalypse it was called.
The AI Agent: your 24/7 employee
It goes beyond standalone tools. The real shift is the AI agent. An AI that works for you 24/7, not just when you ask.
What's the difference with a chatbot? A chatbot waits until you ask something. An AI agent thinks ahead. It checks your calendar, sees you have a presentation tomorrow and prepares an outline. Without you asking. A chatbot tells you how, an AI Agent actually executes it.
> "AI Won't Replace Humans. Humans Who Use AI Will Replace Those Who Don't." > — Sam Altman, OpenAI CEO
The people learning to deploy AI now aren't replaceable. Because a person with the right AI tools delivers the work of an entire team. We may not be there today, but that's where the world is heading. And it's happening faster than most people think.
My tip: pick one task this week. Just one. Something you do every week and that takes too much of your time. Try to automate it with AI.
- Take a paid subscription to Claude or ChatGPT
- Don't use the default model but the best available (Claude Opus 4.6 or GPT-5.3)
- Deploy AI on your hardest tasks, not as a Google replacement
- An hour a day experimenting puts you ahead of 99 percent of people
The gap between people who get AI and people who ignore it grows wider every month.
