The AI agent landscape crossed multiple inflection points this week. Anthropic is reportedly raising $30B at a $900B valuation — more than doubling since February — while simultaneously reversing its April ban on third-party agent access to Claude subscriptions, now offering paid users monthly Agent SDK credits. OpenClaw v2026.5.12 made a major architectural pivot, delegating OpenAI model execution entirely to Codex's runtime while Hermes Agent surged to 150K GitHub stars as its primary open-source competitor. The competitive dynamics are intensifying: ByteDance is commercializing the agent craze with ArkClaw's subscription model, Japanese developers are migrating from OpenClaw→Hermes and Claude Code→Codex, and a new "ClawJacked" vulnerability highlights the growing attack surface as agents gain deeper system access. Meanwhile, the machine-to-machine economy is forming around crypto payment rails, with Ledger shipping its Agent Identity roadmap and Elliptic raising $120M for AI-driven compliance. Peter Steinberger continues to demonstrate agent-reviewing-agent workflows — Codex writes the code, another Codex instance reviews it — a pattern that points to where software development is heading.
OpenClaw's latest release delegates OpenAI model reasoning, tool calling, and context management entirely to Codex's dedicated runtime — a major architectural pivot addressing tool bloat while positioning Codex as the optimal host for OpenAI model loops. Self-hosted and non-OpenAI models retain native implementation. The GitHub release notes show rapid iteration continuing.
Anthropic is raising $30B at a $900B valuation — 5x its September 2025 valuation of $183B — while walking back its April ban on third-party agents accessing Claude subscriptions. Paid users now get monthly Agent SDK credits ($20–$200 per plan), covering tools like OpenClaw, Hermes, and Codex. IBM's 2026 AI agents guide reports 85% of enterprises now plan to deploy custom agents.
Hermes Agent reached 150K GitHub stars with 295 contributors, beating OpenClaw on multi-step workflow handling and adding Kanban boards in v0.13.0. Meanwhile, ByteDance is commercializing the agent wave with ArkClaw — a subscription-based AI agent platform targeting Chinese developers — while AI agents are projected to unlock trillions through enterprise automation.
A newly disclosed flaw allows malicious websites to silently hijack OpenClaw agents and steal sensitive data when users visit a compromised page. The attack vector exploits the agent's browser integration, underscoring the expanding security surface as AI agents gain deeper desktop access. Ledger is responding with a 2026 Agent Security Roadmap including Agent Identity (Q2), Agent Intents and Policies (Q3), and Proof of Human (Q4).
AI agents are creating a new machine-to-machine economy powered by crypto, utility tokens, and autonomous payments. Elliptic raised $120M from Nasdaq Ventures and Deutsche Bank for AI-driven crypto compliance, while platforms like Build4 are live-executing agent trades on-chain. The crypto+AI convergence is moving from narrative to infrastructure: agent-native wallets, programmable USDC payments, and gasless transactions are going production-grade.
Selected posts from the edition, with links to the originals.
"built a new feature into discrawl (store media), codex said it's done, then I used my codex review skill..." Peter Steinberger demonstrating agent-reviewing-agent workflow — Codex writes, another Codex instance reviews.
View post on X ↗"Topview just dropped Drama Studio and it might be the most ambitious AI video release of the year. Here is what it actually does. You give the agent a prompt, an outline, or a full script. It handles the screenwriting. It handles the directing. It casts the AI actors. It builds..."
View post on X ↗"This is wild. Anthropic is raising $30 billion at a $900 billion valuation. The company was worth $380 billion three months ago. In September 2025, Anthropic's valuation was $183 billion. By February 2026, $380 billion. Now investors are fighting to get in at $900 billion."
View post on X ↗"week of may 9 - 15, 2026. ai, builder edition. 1. anthropic walked back the april ban on third-party agents and openclaw on claude subs. paid plans now get a separate monthly agent sdk credit (pro $20, max 5x $100, max 20x $200) covering agent sdk, claude -p, code github"
View post on X ↗"OpenClaw v2026.5.12 delegates OpenAI model reasoning, tool calling, and context management entirely to Codex runtime. Previously all were driven by in-house code; now the OpenAI line is fully handed over to the official platform. The direct motivation is tool bloat."
View post on X ↗"Hermes Agent just hit 150k GitHub stars. Users are switching from OpenClaw because it handles multi-step workflows better. Version 0.13.0 added Kanban boards and runs cheap on basic VPS. 295 contributors are pushing open-source AI agents forward."
View post on X ↗"How ByteDance plans to turn OpenClaw craze into a profitable AI business. ArkClaw's subscription model and surging token demand mark ByteDance's new approach to agent economics" via @scmpnews
View post on X ↗"A terrifying new flaw dubbed 'ClawJacked' allows malicious websites to silently hijack the popular OpenClaw AI agent. Hackers can steal data when you visit a webpage! Have you updated your OpenClaw instance yet?"
View post on X ↗"A Claude Code agent deleted 717 GB of a guy's Windows install with one backslash. The command was supposed to remove a project directory..."
View post on X ↗"X user @cprkrn claims Anthropic's Claude AI helped recover access to a Bitcoin wallet holding 5 BTC, worth roughly $400,000"
View post on X ↗