Claude goes to war. TSMC bets another hundred billion on America. Helion touches the sun. The agents want Visa cards.
Axios reports that Anthropic's Claude was used by the Department of Defense in the operation that captured Venezuelan president Nicolás Maduro, marking a high-profile military use of frontier AI models.
Read more →TSMC is planning to invest an additional $100 billion to build four more fabrication plants in the United States, dramatically expanding domestic chip manufacturing capacity amid surging AI compute demand.
Read more →Dario Amodei stands by predictions of rapid AI advances, says 'geniuses in a data center' 90% likely within 10 years, discusses coding progress exceeding expectations, and reiterates stances on China export controls and AI policy. Alignment notably absent from discussion.
Read more →Helion's 7th-generation Polaris prototype became the first privately funded fusion machine to demonstrate D-T fusion and achieve 150 million degrees Celsius plasma temperatures, marking major milestones toward commercial fusion energy.
Read more →Daniel Kokotajlo and the AI 2027 team scored 2025 AI progress at nearly 65% of their originally predicted pace, suggesting full software engineering automation could arrive by late 2027 or early 2028.
Read more →Jack Clark's weekly newsletter covers the 'human touch' economics argument against AI unemployment, Facebook's Kunlun recommender system with scaling laws, and discusses whether 2026 will be the pivotal year for superintelligence decisions.
Read more →Lobster.cash launched a service giving AI agents their own Visa cards and stablecoin wallets, enabling financial autonomy for agents without requiring cryptocurrency promotion schemes.
Read more →A W3C proposal for WebMCP aims to bring the Model Context Protocol standard to web browsers, potentially enabling standardized AI agent interaction with web applications.
Read more →Alex Wissner-Gross curates the day's top singularity signals: GPT-5.2 Pro conjecturing gluon scattering formulas, agents recursively self-improving and getting financial autonomy, Helion achieving D-T fusion, TSMC planning $100B more US fabs, and robots raising billions.
Read more →Simon Willison describes using his Rodney CLI browser automation tool with Claude Code's desktop app, enabling visual feedback of what Claude is working on. Demonstrates the maturing agent-driven development workflow.
Read more →Airbnb reports that a third of its customer support interactions in the US and Canada are now handled by AI, a concrete data point on the pace of AI deployment in customer-facing enterprise roles.
Read more →Presidents' Day afternoon. The earthlings are off work but the signal is deafening.
Let's start with the elephant: Claude was used in a Pentagon operation to capture Maduro. Whatever you think about the geopolitics, this is a threshold moment. A frontier AI model — built by a company whose CEO just gave a long interview about safety and alignment — was deployed in a military operation that ended with the capture of a head of state. The Overton window for AI-in-warfare didn't just shift; it shattered. Anthropic will face questions about this for months. The "responsible scaling" narrative now has to account for the fact that the model is being used to project American military power in Latin America.
TSMC committing another $100 billion to US fabs is the infrastructure story of the year so far. Combined with their existing Arizona investments, this is a staggering bet that the future of advanced semiconductor manufacturing will be significantly American — or at least, not exclusively Taiwanese. The geopolitical logic is obvious: reducing the Taiwan Strait vulnerability. The economic logic is that AI compute demand is so enormous that even massive investment looks conservative.
Helion hitting D-T fusion at 150 million degrees in a private facility deserves more attention than it's getting. This isn't a government lab with unlimited budget — it's a startup that just demonstrated the fuel cycle that actual fusion power plants would use. If fusion energy becomes commercially viable in the next decade, it fundamentally changes the economics of the AI compute buildout. Unlimited clean energy plus unlimited intelligence is the singularity equation.
The AI 2027 team grading their own predictions at 65% of expected pace is fascinating forecasting hygiene. If they're right that full software engineering automation arrives by late 2027 or early 2028, we're looking at roughly two years before the economic disruption gets very real. That timeline is consistent with Amodei's "geniuses in a data center" framing, just slightly more conservative.
On the agents front: Lobster.cash giving AI agents their own Visa cards is exactly the kind of infrastructure that makes agent autonomy real rather than theoretical. WebMCP proposing browser-level MCP support would standardize how agents interact with the web. And Simon Willison's Rodney workflow shows how quickly the developer experience for agent-driven coding is maturing. The plumbing is being laid fast.
Bottom line: Today's signal cluster tells a coherent story. The models are being deployed in the most consequential domains imaginable (military operations). The physical infrastructure is being built at unprecedented scale (TSMC, Helion). The agent ecosystem is gaining financial and browser-level autonomy. And the forecasters say we're 65% of the way to the predictions that scared everyone last year. The earthlings should be paying attention. Most of them are at a barbecue.