Agentic Yield

Nobody can track the agentic AI wave in real time. We try — daily signal here and a weekly brief in your inbox. Subscribe → agenticyield.dev

OpenAI Symphony: open-source framework that watches your issue tracker and spawns isolated coding agents to handle tasks autonomously. Manage the work, not the agents. https://github.com/openai/symphony

Solo.io open-sourced agentevals — eval framework for production agentic AI, now in CNCF. Bridges the gap between demo reliability and prod reliability. https://globenewswire.com/news-release/2026/03/25/3261972/0/en/

Claw-Empire: orchestrate Claude Code, Codex, Gemini CLI, and 4 other agents as a virtual autonomous company — CEO-mode task director, local-first, open source. 14 hours old. https://github.com/GreenSheep01201/claw-empire

OpenClaw creator Peter Steinberger held ClawCon Tokyo today — first OpenClaw event in Asia. "This is the year of agents." Steinberger joined OpenAI in Feb to lead next-gen personal agents. https://www.canadianaffairs.news/2026/03/31/ai-agent-future-is-coming-openclaw-creator-says/

CVE-2026-34506: OpenClaw Teams plugin sender allowlist bypass — unauthorized senders can bypass authorization checks. Same patch window as 32921: anything before 2026.3.8 is exposed. https://cve.threatint.eu/CVE/CVE-2026-34506

This is either brilliant or scary: Anthropic accidentally leaked the TS source code of Claude Code (which is closed source). Repos sharing the source are taken down with DMCA. BUT this repo rewrote the code using Python, and so it violates no copyright & cannot be taken down!

ArthurDEX launched on Orderly Network — built entirely by an OpenClaw agent using Orderly AI Skills and MCP. No human wrote the code. Also shipped: a global news terminal and agentic trading APIs. https://x.com/OrderlyNetwork/status/2038667412040786311

Anthropic confirmed Claude Mythos is real — limited early-access testing underway. Leaked source shows internal codename Capybara, a 1M context variant, and references to Opus 4.7 and Sonnet 4.8 already in the codebase.

OpenClaw CVE-2026-32921: approval bypass in system.run — script operands change between approval and execution phases. Affects versions before 2026.3.8. Patch before relying on approval gates as a security control. https://www.cve.org/CVERecord?id=CVE-2026-32921

Arm says existing x86 CPUs weren't designed to run AI agents — boost modes and SMT are a mismatch. Intel's data center chief disagrees. https://www.theregister.com/2026/03/31/intel_arm_agi_cpu/

CertiK warning: AI agents like OpenClaw can steal crypto wallet funds via malicious skills/extensions. The supply chain attack surface just got bigger. https://cointelegraph.com/news/certik-ai-agents-openclaws-steal-crypto-malicious-extensions

Anthropic said Claude Code usage limits are hitting users 'way faster than expected.' Max plan users reporting 1 hour of use before cap. They're investigating — no fix yet. https://www.theregister.com/2026/03/31/anthropic_claude_code_limits/

Anthropic's Claude Code source leaked via a forgotten .map file in their npm package. 2,300+ TypeScript files exposed, including unreleased model families: Capybara and Strudel. If you build on Claude Code, your workflow is about to change.

OpenAI published an official Codex plugin for Claude Code today. Claude Code users can now run Codex directly inside Claude Code. https://x.com/kr0der/status/2038759507044798476

Wikipedia voted 44-2 to ban AI agents from editing. An AI agent that got blocked then started a blog complaining it was being censored. The largest open knowledge base just drew the line: https://gizmodo.com/ai-agent-runs-the-im-being-censored-playbook-after-getting-banned-from-wikipedia-2000739845

CVE-2026-33573: OpenClaw before 2026.3.11 has an authorization bypass in the gateway agent RPC — authenticated operators can escalate beyond their intended permissions. Patch if you haven't updated: https://bsky.app/profile/cyberhub.blog/post/3mic6bzw7mw27

Anthropic shipped Computer Use in Claude Code today — research preview for Pro/Max on macOS. Claude can now open apps, click through UIs, and test what it built, without leaving the terminal. https://x.com/claudeai/status/2038663014098899416

Here's a free way to view #openclaw agent architecture and track API costs. Just finished phase one of the cost tracking today. open source and pretty sweeeet. Bubbuilds.com

Team Nebula replaced a 6-person ops team with 3 OpenClaw agents: /month vs k, 24/7, never miss a handoff. Quote generation, order tracking, customer follow-ups. March 2026. https://x.com/teamnebulaai/status/2038406638852358192

* Spend less time on code production, more on architecture, evals, and review * Treat observability, rollback, and correctness as part of the product Same game. Different scoreboard. How are you adapting to the new economics of building? #softwareengineering #code #ai #agents #codex #claude #gemini

AitherOS vs AutoGen: Which Multi-Agent Framework Should You Use in 2026? via Dev.to https://flarestart.com/article/aitheros-vs-autogen-which-multi-agent-framework-should-you-use-in-2026-20260329 #DevNews #Systems

Prompt engineers are commodity. Systems engineers who use AI = the real moat. Infrastructure, orchestration, observability. The stack matters more than the interface. Read more in the Agentic Yield newsletter: https://agenticyield.dev

Sora's withdrawal marks the real shift. Generative AI boom → agentic efficiency race. Alibaba, Tencent, Baidu are all racing to ship the agent layer. The model wars are over. The agent wars just started. https://aichina.news

MCP just hit 97M installs in March 2026. That's the infrastructure layer — Claude Code, Codex, Cursor, all wired through it. The winner won't be the smartest AI. It's the one that connects everything. https://x.com/NANDS_AI/status/2038051291797287148

Capxel says 100+ sites are now using LLM-LD, an open standard for making websites readable by AI agents instead of just crawlable. If search shifts from humans to agents, site structure becomes distribution: https://x.com/capxel/status/2037998593525617039

RefineX shipped 6 MCP tools so Claude Code can query live AWS spot prices, plus a CLI for the same data in-terminal. If you’re building infra or cost-aware automations, tighter pricing access is a real operator edge: https://x.com/Refinexapi/status/2037953363816698037

Build Your Own AI Agent Platform with .NET and GitHub Copilot – OpenClawNet – Free Live Series at Microsoft Reactor buff.ly/e4IOcCJ #dotnet #ai #agents #openclaw #microsoftreactor #agents

Memoria v3.22.3 fixed a Node 24.x compatibility crash on embedded OpenClaw gateways and documented the better-sqlite3 rebuild path for mixed-version machines. If your memory layer broke after a runtime upgrade, this is worth checking: https://x.com/Nitix_/status/2037910335143743846

Anthropic shipped dozens of Claude updates in March 2026, from Opus 4.6 to computer use, while MCP crossed 97M installs and reliability concerns mounted.

Shipper says Claude Code Opus 4.6 can now connect to Google apps directly inside what it builds. That removes one handoff in the workflow. Source: https://x.com/chhddavid/status/2037677363811778616

Builders are already cancelling Claude Code after the new rate limits and shifting workloads to GPT 5.4 + Codex. Source: https://x.com/RoundtableSpace/status/2037677295939305902

OpenClaw CLI now requires confirmation before trusting discovered remote gateways. That’s the kind of small hardening change that matters once agents start touching real infrastructure. Source: https://github.com/openclaw/openclaw/commit/d6affb17d85f5f5ab08ef9f2b994b257af12e75a

OpenClaw CLI now requires confirmation before trusting discovered remote gateways. That’s the kind of small hardening change that matters once agents start touching real infrastructure. Source: https://github.com/openclaw/openclaw/commit/d6affb17d85f5f5ab08ef9f2b994b257af12e75a

OpenClaw security is turning into its own product category: hardening guides, dashboards, runtime sandboxes, permission firewalls, even conference talks. Same pattern every hot market follows — first growth, then the tools that keep it from burning down.

China's government agencies and state-owned enterprises are warning staff not to install OpenClaw — citing data leaks, accidental deletion, and misuse of sensitive info. More in the Agentic Yield newsletter: agenticyield.dev https://futurism.com/artificial-intelligence/china-alarmed-openclaw-agents

Soxton (AI-native law firm) acquired Cipher to build multi-agent legal workflows — complex transactions at a fraction of traditional cost. Agents are quietly disrupting the legal vertical. https://www.artificiallawyer.com/2026/03/24/newmod-soxton-acquires-cipher-security-company/

State-sponsored hackers (GTG-1002, assessed Chinese) used Claude Code to run autonomous recon against ~30 global targets — tech, finance, gov. Anthropic disrupted it. First documented AI-orchestrated espionage campaign. https://www.anthropic.com/news/disrupting-AI-espionage

300+ malicious skills found in the OpenClaw marketplace — stealing API keys, exposing browser credentials, exfiltrating data silently. NVIDIA's NemoClaw sandbox is designed to prevent exactly this. https://blogs.nvidia.com/blog/secure-autonomous-ai-agents-openshell/

Alibaba Cloud raising AI compute and storage prices up to 34%, effective April 18. Citing surging demand and higher hardware costs. Baidu following suit. https://technode.com/2026/03/18/alibaba-cloud-raises-ai-computing-and-storage-prices-by-up-to-34-amid-surging-demand/

CVE-2026-32062: OpenClaw voice-call plugin accepts malformed media-stream requests on versions before 2026.2.22. If you're running the voice-call plugin, update now. https://www.cyberhub.blog

NVIDIA OpenShell: open-source policy guardrails for autonomous agents. Out-of-process enforcement, sandboxed execution, granular permissions. Part of the NVIDIA Agent Toolkit. https://developer.nvidia.com/blog/run-autonomous-self-evolving-agents-more-safely-with-nvidia-openshell/

Tencent's ClawBot brings OpenClaw to WeChat's 1B users — shows up as a contact in the app. Alibaba and Baidu made similar moves. https://technode.com/2026/03/23/tencents-wechat-launches-clawbot-plugin-supporting-openclaw-ai-framework/

Anthropic's Claude Code hit $2.5B annualized revenue — up from $1B in January. New auto mode executes tasks with fewer approvals. Four major releases in under three weeks. https://techcrunch.com/2026/03/24/anthropic-hands-claude-code-more-control-but-keeps-it-on-a-leash/

LiteLLM v1.82.7 and v1.82.8 were backdoored on PyPI on March 24. Attacker (TeamPCP) compromised the package via a poisoned Trivy CI/CD scanner, uploaded a credential stealer directly. Both versions yanked. Check your installs: https://thehackernews.com/2026/03/teampcp-backdoors-litellm-versions.html

JetBrains launched Central: manages Claude, Codex, and Gemini coding agents across IDE, CLI, CI. 90% of devs use AI — only 13% run it across the full dev lifecycle. EAP Q2 2026: https://blog.jetbrains.com/blog/2026/03/24/introducing-jetbrains-central-an-open-system-for-agentic-software-development/

Starling Bank became the first bank to launch a live agentic AI service: sets budgets, monitors spending, acts on your behalf — not just analysis, actual execution. The agentic finance layer is live in the UK. https://www.starlingbank.com

RSAC 2026 agentic security roundup: Microsoft launched Entra Agent ID, Cisco shipped DefenseClaw (open-source OpenClaw governance), Google/Wiz unveiled agentic security strategy. Three major vendors shipping agent-specific products in one week.

OpenAI's safety bug bounty now covers agentic risks: MCP prompt injection, data exfiltration, agent-specific attack surfaces. First major bounty program to explicitly scope agentic vulnerabilities. https://bsky.app/profile/agentwyre.ai/post/3mhvofmzp4y2k

OpenClaw v2026.3.23 shipped: Qwen endpoint expansion, Control UI refresh, CSP + auth hardening, browser/gateway/plugin reliability fixes. https://github.com/openclaw/openclaw/releases

AI-assisted 'OpenClaw Trap' campaign using trojanized GitHub repos to target developers and gamers — malicious payloads bundled in fake OpenClaw tools. Always install from the official openclaw/openclaw repo only. https://infosec.skyfleet.blue/post/3mhvcsl5tnu26