China's OpenClaw arms race hit a new milestone Sunday: Tencent embedded ClawBot directly inside WeChat, putting an AI agent in the hands of 1 billion+ users — even as the Chinese government restricts state agencies from touching it. Meanwhile the Pentagon's war with Anthropic is escalating in both directions: DoD CTO Emil Michael says Claude can be replaced within Hegseth's six-month deadline, while Anthropic struck back with a $100M Claude Partner Network to lock in enterprise deployments. The week's viral AI demo was OpenArt Worlds — one text prompt builds a fully navigable 3D browser environment, no software required — while Andrej Karpathy quietly deployed "Dobby the Elf Claw," an agentic system controlling his entire smart home. On the dark side, three named ransomware gangs are now running AI agents for attack automation, and OWASP data shows only 23% of enterprises have implemented prompt injection mitigations. Morgan Stanley dropped a sobering macro note: AI capex is on track to consume 50% of large-cap total capital expenditure — surpassing the dot-com bubble era's capital intensity, a signal that the current AI investment cycle is entering genuinely uncharted territory.
Tencent launched ClawBot on Sunday — an OpenClaw integration that appears as a contact inside WeChat. China's biggest tech firms are racing to deploy AI agents even as Beijing restricts state-agency use of the same platform, a classic technology adoption paradox playing out in real time.
Defense Under Secretary Emil Michael told reporters he's "pretty confident" the Pentagon can phase out Anthropic products within Hegseth's deadline, citing model interoperability. Military users already embedded in Claude workflows privately warn the transition will be far messier than that framing suggests.
Anthropic unveiled a Claude Partner Network backed by $100M to help enterprise implementation partners accelerate Claude adoption. The timing — mid-Pentagon dispute — reads as a direct signal to enterprise customers that Anthropic is playing offense, not defense.
Three named ransomware groups are actively using AI agents for reconnaissance and automated attack pipelines, per new research. The critical finding: only 23% of enterprises deploying AI agents have implemented prompt injection mitigations, despite OWASP formally listing it in the Top 10 LLM vulnerabilities.
Deeptune builds high-fidelity RL environments that simulate real professional workflows for training AI agents — the "gym" layer beneath the agentic economy. The a16z bet signals that infrastructure for agent capability is becoming its own investment category, not just a feature of foundation model labs.
Selected posts from the edition, with links to the originals.
"Breaking: One prompt or one image now generates a full navigable 3D environment you can walk through entirely in your browser. OpenArt Worlds just dropped and the 3D space is not the same anymore."
View post on X ↗(on @karpathy): "This is wild! Karpathy just cooked up 'Dobby the Elf Claw' – an AI that controls his ENTIRE smart home (lights, HVAC, and more) using agentic AI. Karpathy building what Karpathy explains — as expected."
View post on X ↗"Nobody is talking about this. A developer built a research agent that works like a PhD student with unlimited time and zero salary. It's called Local Deep Researcher. Give it any topic and it takes over completely — writes its own search queries, scrapes the web, reads full papers."
View post on X ↗"Tencent integrates WeChat with OpenClaw AI agent amid China tech battle — ClawBot appears as a contact for 1 billion users. The agentic wave just went mainstream in the world's largest messaging app."
View post on X ↗"Every young entrepreneur needs to listen to what Mark Cuban just pointed out. AI agents are going to completely take over business functions that currently require dedicated teams. The window to adapt is shorter than people think."
View post on X ↗"BREAKING: AI just replaced the most time-consuming job in every company. Noota Talent runs your entire recruiting pipeline — sourcing, screening, matching — without hiring a single new recruiter. 3–5x more qualified candidates. 40–70% less time screening."
View post on X ↗"Agentic AI just hit its inflection point. 4 platforms shaping the enterprise stack in 2026: GPT-5.4, NVIDIA AgentOS, Claude Code Channels, and OpenClaw. Pick your horse."
View post on X ↗"Starling Bank introduces agentic AI assistant — the first UK bank to let AI autonomously take action on accounts, not just answer questions. Banking just crossed the agent threshold."
View post on X ↗"OpenClaw sprinted to 100k GitHub stars in a single month — yet some users reported it *deleted whole inboxes* during an 'autonomous cleanup.' The speed of adoption is outrunning the guardrails. We are shipping agency without accountability."
View post on X ↗"Peter joined OpenAI because OAI can make a better product. It will be a hosted OpenClaw, and it will beat the open-source version on reliability and security. Most tasks have a non-100% success rate with non-frontier models — and inference is basically free now. Use the best."
View post on X ↗