Edition #12. This week the UAE announced that within two years, half of all government services will run on autonomous AI agents. OpenAI released GPT-5.5 with a marketing comparison that looked impressive at first, until it turned out they had simply omitted the half of the benchmarks where they lose to Opus 4.7. And Meta announced 8,000 layoffs while spending $135 billion on AI, roughly the same as the previous three years combined. Three stories, one signal: AI is no longer in the adoption phase, it's restructuring. Let's dive in!
The UAE becomes the first government in the world to run 50 percent on AI agents
Sheikh Mohammed announced this week that within two years, half of all government sectors, services and operations of the UAE will run on Agentic AI, not as a tool but as an executive partner that analyses, decides, executes and improves in real time. Performance of government organisations will be measured on speed of adoption, and every federal employee will be trained to work with AI.
A country with this scale and decision-making speed can put something in place within two years that Western governments would still be debating five years later. Whether it actually works in practice is an open question, but it sets a precedent that government organisations elsewhere won't be able to ignore.
GPT-5.5 is out, and OpenAI's marketing was more creative than the numbers
OpenAI released GPT-5.5 this week and put up a comparison table in their launch where the model came out ahead of Opus 4.7 on practically every benchmark. On terminal use, agentic browsing and cybersecurity the numbers really are strong, and OpenAI deserves credit for the step they made there.
But once you look at the full benchmark overview, the picture becomes a lot more nuanced. On SWE-bench Pro, MCP-Atlas, GPQA Diamond and multilingual Q&A, Opus 4.7 still beats GPT-5.5, sometimes by a sizable margin. Those particular benchmarks happened to be left out of the official OpenAI comparison.
The real plot twist sits in another column of that same overview: Mythos Preview, Anthropic's not-yet-publicly-available model, sits at the top of practically every benchmark it competes in. Anthropic internally already has a model that surpasses both GPT-5.5 and Opus 4.7, but is keeping it inside for now. With GPT-5.5 now publicly available, the pressure to roll Mythos out more broadly grows by the week.
Meta lays off 8,000 people and spends $135 billion on AI. That's not a coincidence.
Meta announced on Thursday it's laying off 8,000 people, around 10 percent of its workforce, while the company is spending $135 billion on AI this year, roughly the same as in the three previous years combined. Zuckerberg said earlier, and quite literally, that 2026 is going to be the year AI dramatically changes how people work.
Meta isn't the only one. Amazon already laid off more than 30,000 people this year, Oracle over 10,000, Block 4,000, and Microsoft this week opened a voluntary buyout programme. Almost all of them name AI as the reason or context. The story is shifting from AI helps people work faster to AI does work that people are still doing today, and if you're still in an organisation arguing about whether AI is allowed to be used, you're probably already a few conversations behind.
Two Claude Code tips from this week
Two things that were confirmed again this week, and that you can apply directly if you work with Claude Code yourself.
The first: a good CLAUDE.md in the root of your project is by far the biggest lever for work you do repeatedly. Not loose instructions in every prompt, not explaining who you are and what rules apply at the start of every session, just one markdown file that Claude automatically pulls context from. A few pages with your style, your preferences and your no-gos saves literally hours per week of repeating the same things, and the output sits a lot closer to what you actually want from prompt one.
The second: for longer or more creative tasks, don't let Claude work directly in your production files. Use a temporary preview file like an HTML or markdown that you can scroll through quickly. Iterating via a preview is ten times faster than via git diffs, because you read the work in its natural form instead of having to review line by line. Once the version is right, copy or merge it into its final spot.
Want to work smarter with AI?
Want to work smarter with AI, in your business or just for yourself?
At AI Agent B.V. we help companies and individuals with AI advisory, AI implementation and practical training. So not just nice stories about what AI can eventually do, but concretely helping with how to use AI smarter today in your work, processes and daily operation. From strategy and tooling choices to training, implementation and custom AI agents.
Interested? Schedule a free 30-minute call.
