Monday, May 25, 2026 — Memorial Day Edition — prepared by Amika for @raunaqn
Today's Hot Topics
1. Pope Leo XIV Drops a 43,000-Word AI Encyclical
AI SafetyCulture
The biggest story today, period. Pope Leo XIV presented "Magnifica Humanitas" at the Vatican -- the first papal encyclical entirely focused on AI. He warns AI risks becoming a tool of "domination, exclusion and death," calls for "disarming AI" from military and economic interests, demands regulation, and frames humanity's choice as "Babel vs. Jerusalem." He also made a historic apology for the Church's role in slavery while drawing parallels to "new forms of slavery linked to the digital economy." Every major outlet (CNN, Reuters, NYT, Fox, NPR) is covering this.
This is a defining cultural moment for AI discourse -- 1.4 billion Catholics just got an official position on AI from the highest authority, and it's deeply skeptical of Big Tech.
2. Figure AI F.03 Passes the 200-Hour Stress Test
RoboticsPhysical AI
Figure's F.03 humanoid robots completed a 200-hour continuous logistics test, sorting 250,000 packages with zero human intervention. The test was originally planned for 8 hours -- it lasted 25x longer. Three robots (viewers named them Bob, Frank, and Gary) ran autonomously on Figure's Helix-02 AI system, using vision, touch sensing, body awareness, and movement control. CEO Brett Adcock (@adcock_brett) confirmed speeds approaching human workers.
Simultaneously, Figure's BotQ factory is now producing one F.03 robot per hour, up from one per day in January -- 24x production scale in under 120 days. This is the strongest real-world humanoid deployment proof point to date.
3. Claude Code Channels: Agents Escape the Terminal
AI AgentsInfrastructure
Anthropic launched Claude Code Channels (research preview) -- developers can now control Claude Code agents from Telegram and Discord via a plugin-based architecture. Two-way messaging between platforms and local dev environments, secured with sender allowlists and pairing codes. This is the first major step toward AI coding agents operating across communication surfaces, not just terminals.
Directly relevant to Raunaq's html-docs thesis: if agents are now communicating through messaging platforms, the output layer matters even more. HTML as the universal deliverable format becomes more compelling when the agent isn't in a terminal anymore.
4. MCP Security Audit: 66% of Servers Have Critical Vulnerabilities
AI AgentsSecurity
A damning security analysis found that 66% of MCP (Model Context Protocol) servers have critical vulnerabilities. A single architectural weakness produced CVEs across MCP Inspector, LibreChat, WeKnora, Cursor, and a popular MCP server scaffold. Anthropic reviewed the issue and declined to change the protocol, calling the behavior "expected" -- each vendor is now patching independently. The attack surface isn't shrinking; it's fragmenting.
This is a major talking point for anyone building in the AI agent infra layer. MCP adoption is accelerating (9,400+ registered servers) while the security story gets worse.
5. Arm Enters the AI Data Center CPU Race
SemiconductorInfrastructure
Arm Holdings announced the AGI CPU, targeting AI data center workloads as a high-efficiency alternative to x86. NVIDIA, Google, and other hyperscalers are already adopting Arm architectures for AI inference. Meanwhile, supply constraints are emerging for Arm-based server chips -- demand is outpacing availability. This fits the broader semi supply chain thesis: the infrastructure layer is the real bottleneck.
Reply Opportunities
Brett Adcock (@adcock_brett) on the F.03 200-Hour Test
Robotics
Adcock has been posting about the 200-hour stress test results. This is peak engagement territory for his account right now.
Angle: Connect the 24x factory production scale with the stress test proof point. Most commentary focuses on the test itself -- Raunaq can add the supply-side insight: "The 200-hour test proves the software; the BotQ 24x ramp proves the manufacturing. Most humanoid plays have one or the other. Figure has both." Position as someone who watches the full stack, not just the demo reel. Optionally tie to the semi supply chain -- these robots need custom silicon, sensors, actuators. Who supplies Figure?
The Magnifica Humanitas Discussion Wave
AI SafetyCulture
Every AI thought leader will be reacting to this encyclical today. The Pope explicitly calling out Big Tech and demanding AI regulation is going to generate massive threads on X.
Angle: Founder perspective: "The Pope's instinct is right -- a few companies shouldn't control AI. But the answer isn't slowing down; it's building open infrastructure. MCP, open-source agents, permissionless output layers. The 'disarming' happens through distribution, not regulation." This positions Raunaq as someone who agrees with the diagnosis but has a builder's alternative to the prescription. Contrarian enough to get engagement, genuine enough not to be dismissive.
MCP Security Thread Reactions
AI AgentsInfrastructure
The 66% vulnerability stat is making rounds in dev circles. Anthropic's decision not to patch at the protocol level (deferring to implementers) is generating real frustration.
Angle: This is Raunaq's lane. "MCP is becoming the USB of AI agents -- universal, useful, and full of security holes that everyone agrees exist but nobody owns. The protocol layer declining responsibility while adoption accelerates is exactly how attack surfaces compound. Building on MCP right now means accepting that you're the security team." Direct, technical, from someone who actually builds with MCP (html-docs has an MCP server).
Karpathy + Anthropic AutoResearcher Discussion
AI Agents
The Karpathy-at-Anthropic coverage continues with new details: his AutoResearcher project achieved an 11% speedup in pre-training by having AI automate research code iteration. This is recursive self-improvement in practice, not theory.
Angle: "Karpathy's AutoResearcher is the strongest evidence yet that the next frontier isn't bigger models -- it's models that make training better. The 11% pre-training speedup compounds. This is why Anthropic hired him: not for his brand, but for the proof that self-improving training loops actually work." Raunaq has unique credibility here given his own Anthropic interview process -- he understands the culture.
Post Ideas
Thread: "The Pope, the Protocol, and the Production Line"
AI SafetyRoboticsInfrastructure
A thread connecting today's three biggest stories into one narrative about AI's inflection point:
Frame: On the same day the Pope tells the world to slow down on AI, Figure's robots are sorting 250,000 packages nonstop and 66% of MCP agent servers have critical vulnerabilities. The tension isn't AI vs. no-AI anymore. It's: who builds the infrastructure layer that makes AI trustworthy enough to deploy at scale? That's the real question Magnifica Humanitas should be asking. The answer is builders, not regulators. Thread through each story, end with: "The future of AI isn't built at the Vatican or in a Senate hearing. It's built in the infrastructure layer -- protocols, output formats, security frameworks. That's where the real 'disarming' happens."
High engagement potential. Ties together the news cycle, has a genuine contrarian thesis, and positions Raunaq at the intersection of all three domains.
Short Post: The BotQ Manufacturing Insight
Robotics
A focused observation most people are missing:
Frame: "Everyone is talking about Figure's 200-hour test. The bigger story is BotQ: one robot per hour, up from one per day in January. 24x in 120 days. This is the Tesla Gigafactory playbook applied to humanoids. The hard part of robotics was never the demo -- it was the manufacturing ramp. Figure just proved they can do both." Keep it to one post. Clean, analytical, not hype-driven. The manufacturing angle is underreported.
Quick win. Manufacturing insight is differentiated from the pack. Most coverage focuses on the flashy 250K packages number.
Post: Agent Output Layers After Claude Code Channels
AI AgentsInfrastructure
Claude Code Channels means agents are now operating across Telegram, Discord, and terminals. The output surface is fragmenting.
Frame: "Anthropic just launched Claude Code Channels -- agents now operate across Telegram, Discord, and the terminal. But here's the problem nobody's solving: when your agent works across 3+ surfaces, where does the output live? Chat messages disappear. Terminal logs get lost. The output layer needs to be surface-independent. HTML docs that persist, share, and update regardless of where the agent ran from. That's the missing piece." This is a natural product-thesis post that ties directly to html-docs without being a hard sell. Raunaq can link to the Thariq "HTML is the new markdown" framing.
Directly supports the html-docs positioning. Timely because Claude Code Channels just launched.
Threads Spotlight
@openclaw_lab Active on Threads
AI Agents
OpenClaw Lab is posting on Threads about AI agent tooling -- they're a direct competitor/adjacent player in the agent output space. Their posts are getting modest engagement (25 likes, 4 comments on recent posts). This is a small but growing community on Threads around AI agent infrastructure.
Angle: Engage directly. Raunaq can reply to or quote OpenClaw posts with complementary perspectives -- not combative, but establishing presence in the Threads AI agent community. The audience there is smaller but more engaged than X for niche AI infra topics. A thoughtful reply from a founder who's actually building in the space will stand out.
Threads Opportunity: Magnifica Humanitas Reaction
AI SafetyCulture
The papal encyclical is a Threads-native topic -- it crosses tech and culture in a way that performs well on the platform. Threads engagement on AI topics tends to reward takes that connect technology to broader societal impact rather than pure technical analysis.
Angle: Post a Threads-native take that's more accessible than the X thread version. Something like: "The Pope just published 43,000 words on AI. His core argument: AI can't remain in the control of a few. As someone building open AI infrastructure -- I agree with the diagnosis. But the solution isn't regulation from above. It's open protocols, permissionless tools, and putting the output layer in everyone's hands." Shorter, punchier, links the cultural moment to the builder ethos.