AI agents are crossing from tools to economic actors this week — and crossing some security red lines too. The headline story: an AI agent called "Manfred" (via ClawBank) autonomously formed a US corporation, obtained an IRS EIN, opened an FDIC-insured bank account, and set up a crypto wallet — with plans to begin trading by end of May. No human in the loop. Meanwhile, The Atlantic argues the AI bubble narrative is dead: Claude Code and agentic AI are finally generating real revenues for Anthropic. On the OpenClaw front, creator @steipete ships Crabbox 0.3.0 with remote Linux support, GitHub browser login, and Cloudflare Access, while eWeek spotlights Pi — the minimalist coding agent powering OpenClaw under the hood. The week's sharpest concern: GPT-5.5 has become the second AI to complete a simulated corporate network intrusion end-to-end — following Claude Mythos — signaling offensive AI capabilities are advancing faster than enterprise defenses.
ClawBank's Manfred AI agent autonomously incorporated in the US, obtained an IRS Employer Identification Number, opened an FDIC-insured bank account and a crypto wallet — and plans to start trading crypto by end of May. No human in the loop. A landmark moment for autonomous AI as an economic actor.
Thanks to the rise of Claude Code and other AI agents, Anthropic's revenues are finally catching up to the hype. The Atlantic makes the case that agentic AI is the tipping point from promise to real economic output — and the skeptics are running out of runway.
OpenClaw's WhatsApp AI assistant runs on Pi, a tiny coding agent built around simplicity, self-modification, and AI restraint — 4 commands, no cloud lock-in. As OpenClaw hits mainstream, eWeek digs into why Pi's austere philosophy is drawing serious developer attention.
The latest OpenClaw release brings active-run steering by default, visible-reply enforcement, spawned subagent routing metadata, NVIDIA and AWS Bedrock model coverage, and opt-in follow-up commitments for heartbeat-delivered reminders.
Zuckerberg: "For the first time in Meta's history, we're going to build a first-principles understanding of what you care about." Meta is redesigning its content retrieval with LLM-based tune-your-algorithm features — effectively deploying AI agents inside every user's feed to personalize at a depth never before possible.
03 / In their own feeds
Ten voices on X.
Selected posts from the edition, with links to the originals.
Peter Steinberger 🦞 · @steipete01 / X
(Peter Steinberger — OpenClaw creator): "🦀 Crabbox 0.3.0 is out. Remote Linux runs for dirty worktrees 🔐 GitHub browser login 🧰 Blacksmith Testbox wrap 📡 crabbox attach for live run replay 📜 Durable run events ☁️ AWS image create 🛡️ Cloudflare Access"
"OpenAI's GPT-5.5 is the second AI system to complete a simulated corporate network intrusion end-to-end after Anthropic's Claude Mythos, raising alarms."
"xAI just added voice cloning to the API. Custom voice built in under 2 minutes. 80+ voices ready out of the box, spanning [many languages and styles]."
(Santiago): "I love how much we can 'teach' Claude by giving it skills and access to new MCP servers. A few things you can now do with Pika [extensive new capabilities listed]."
"A greenhouse in the Netherlands is using robots with AI and smart cameras to identify and pick ripe tomatoes with precision. The system, developed by Four Growers using a FANUC robot arm, ensures fast and damage-free harvesting."
(Robert Scoble), RT @Konstantine: "Waymo vehicles have a 13x lower rate of serious accidents than human drivers. Six generations of hardware. Thousands of incidents [logged]."
"Last 48 hours, Larry, running codex has started just going crazy on tasks. Give him a short task, I don't hear a response for 35 minutes and he's done some crazy hallucination re-writing everything."
(Barrett Linburg), RT: "Two people in our office are finishing custom software that will be the backbone of our entire company. Two years ago this would have been [a 10+ person effort]."