Anthropic launches Claude Duet to chase Meta's Muse, while a DeepMind engineer resigns over Pentagon AI work. Plus Stripe opens AI-agent banking, Wall Street's AI debt pile grows, Stratechery on SpaceX and bubbles, Pragmatic Engineer on a day of AI-linked security incidents, Sean Goedecke on why laypeople back technical fields, Simon Willison's agent-built RTS game, a skeptic's take on agent-swarm superintelligence, and a wildcard: a rock that got colder the deeper it sank.

TechCrunch · Oct 22

Anthropic launched Claude Duet, a proactive Opus 5.5 agent that messages users first and acts with approval.

Jack Hodgkin, WebProNews · Oct 21

Stripe's new Treasury for agents lets AI assistants check balances, pay invoices, and move funds.

Sam Biddle, The Intercept · Oct 21

A DeepMind research engineer resigned over Google's classified Pentagon and Israeli military AI contracts.

Jack McGeever, Daily Caller · Oct 21

Analysis finds AI firms raised $466 billion in debt this year, with off-balance-sheet commitments mounting.

Ben Thompson, Stratechery · Oct 22

Interview with analyst Craig Moffett comparing today's AI infrastructure spending to the dot-com bubble.

Gergely Orosz, The Pragmatic Engineer · Oct 22

Weekly roundup covers a cluster of AI-linked security breaches and whether AI is changing security practice.

Sean Goedecke · Oct 20

Essay argues people value programming for its practical output, not as art, complicating anti-AI framing.

Simon Willison · Oct 20

Willison built a browser real-time strategy game via 321 AI-agent commits, with audio-synced local multiplayer.

Ramez Naam, Noahpinion · Oct 19

Guest essay argues multi-agent scaling shows diminishing returns, countering claims of agent-swarm superintelligence.