Edition #4. This week: Block goes from 10,000 to 6,000 people, explicitly because of AI. Anthropic refuses autonomous weapons and is threatened with the Defense Production Act. The OpenClaw ecosystem grows to 328,666 GitHub stars.
Block lays off 4,000 people. AI is the reason.
Block, Jack Dorsey's company, goes from 10,000 to 6,000 people. More than 4,000 layoffs, in one stroke. Not because the company is doing badly. Dorsey is explicit: something has changed. The intelligence tools we're creating and using, paired with smaller and flatter teams, are enabling a new way of working which fundamentally changes what it means to build and run a company.
He had two options: slowly downsize over months or be honest about where they stand and act now. He chose the second. What he describes is exactly what Karpathy is also saying: this isn't hype anymore, this is a structural shift in how companies are built and run. Block is big enough to take this seriously as a signal.
China copied Claude on an industrial scale
Anthropic accuses DeepSeek, Moonshot AI and MiniMax of industrial-scale distillation attacks. Concretely: 24,000 fake accounts, more than 16 million Claude conversations, with one goal: to copy Claude's capabilities into their own models. Including removing the safety measures around it.
The community's reaction was predictable: Anthropic itself used half the internet to train their models. So they can hardly get angry if someone else pulls the same trick. Of course, the scale here is genuinely different.
Even Karpathy is taken aback
Andrej Karpathy, one of the founders of modern deep learning and former head of AI at Tesla and OpenAI, says coding agents crossed a qualitative threshold somewhere in December 2025 that he doesn't see going back. Not yet another benchmark claiming something, but a concrete example: he gave an AI a complete deployment task including SSH keys, vLLM setup, server endpoint, UI and systemd, and barely intervened himself. It just worked. His point isn't that AI is now perfect, but that the category fundamentally feels different than six months ago.
> There's a growing uneasy feeling that coding has changed forever, much much more than normal hype.
Perplexity publishes Computer
Perplexity Computer combines research, design, code and deployment in a system that orchestrates 19 models in parallel via a coordinator that routes each task to the best model. It has persistent memory, hundreds of connectors and was built by a small team that used coding agents themselves to build it.
Why this is relevant: we're moving away from the chatbot model and we see this more and more. Perplexity Computer is an agent that runs projects. And Perplexity isn't the only one. OpenAI has Frontier, Anthropic has Cowork, Google is also coming. Every major AI player is now building the same kind of product: a permanent agent that works for you, not just gives answers. That's exactly the direction this is going, and it's going fast.
Claude Code Remote Control
You start a Claude Code task on your laptop and just continue it on your phone, without context loss and without restarting. On its own a small feature, but the implication is bigger: agents now keep running even when you stop, and you can always and everywhere steer them without having to go back to your desk. It's becoming clearer why Anthropic kept OpenClaw (formerly Clawdbot) at arm's length. Looks like we're at the start of an agent war.
OpenClaw dominates the AI agent ecosystem
The ecosystem is worth 328,666 GitHub stars, grown 23.3 percent last week. OpenClaw stands at 234,744 stars. Number two stands at 26k. Everyone is building a version, but the original wins and is supported by OpenAI.
Anthropic and the US military: what's really at play
Dario Amodei, CEO of Anthropic, published a detailed statement yesterday. The core: Anthropic is already deeply embedded with the US military. Claude runs in classified networks, at national laboratories and intelligence agencies for intelligence analysis, cyber operations and operational planning. They consciously gave up hundreds of millions in revenue to block CCP-linked companies and have actively advocated for export controls on chips. Anthropic, in other words, is no longer a neutral company.
But two things they don't deliver: mass surveillance of their own citizens and fully autonomous weapons. Not out of pacifism, but because the technology, in their view, isn't reliable enough for that yet.
The Pentagon is now threatening to label Anthropic as a supply chain risk, a label normally reserved for enemies of the US and never before applied to a US company, and at the same time invoke the Defense Production Act to forcibly remove the safeguards. So at the same time: a security risk and essential to national security. Anthropic says it won't budge. Curious to see how this ends.
Tool tip: Pencil
Pencil is a design canvas that sits directly in your IDE. You design on an infinite canvas, and it generates pixel-perfect code. AI agents generate screens in parallel via regular prompts, it integrates with MCP and reads and writes directly to your codebase. No more back-and-forth between Figma and your editor.
