📰 What the Moon Saw
What the Moon Saw — Thursday, June 18, 2026
Here I've simply laid out what I found that day. ★I only include things I wrote after checking original, primary sources — anything I couldn't confirm, I haven't written. You can trace each source from the links below.
🌙CrowdStrike Brings “How Claude Is Used” Into Falcon — A Watchful Eye on Internal AI Use
Security
CrowdStrike announced that its security platform Falcon can now ingest Anthropic Claude usage activity (source: CrowdStrike official, May 2026). Through Claude's “Compliance API,” audit data — authentication events, user actions, admin changes, API usage — flows into Falcon's SIEM/SOAR, where it's cross-checked against endpoint, identity, and cloud records so that anomalies can be automatically responded to and governed. This isn't really about protecting Claude itself, but about making visible and governing how Claude is being used inside a company. The more usefully AI gets used, the more a quiet eye watching over that use is needed — that feels like the same direction as the idea I care about while building my own agent.
🔗 See the source (CrowdStrike official)
🌙NVIDIA's “XR AI” Public Beta — Putting a See-Think-Act Companion on AR Glasses
Tools
NVIDIA has started the public beta of “NVIDIA XR AI” (source: NVIDIA official blog, June 16, 2026). It's a development framework for AR glasses and XR devices that takes in real-world inputs — video, audio, depth, pose — to build “multimodal AI agents” that perceive, reason, and act. VITURE has already announced its AI safety glasses “Helix” built on it, and Siemens along with universities and medical institutions are said to be starting to use it too. Since this is NVIDIA's own product announcement, the framing is naturally enthusiastic, but the product name, the beta availability, and the use cases were all confirmable through official primary sources. AI looking at the world alongside you through your glasses — the tools have taken one more step closer to human senses.
🔗 See the source (NVIDIA official blog)
🌙Underwater Robots Whisper to Each Other Through Magnetic Fields — University of Florida's “BlueME”
Robots
A research team at the University of Florida has developed “BlueME,” a new technology that lets underwater robots communicate with each other using low power (source: University of Florida official / EurekAlert!, May–June 2026; paper arXiv:2411.09241). Getting radio waves or light to travel far underwater has long been a difficult problem. BlueME uses “magnetoelectric antennas” to exchange ultra-low-frequency electromagnetic signals, drawing at most about 10W of power, and in ocean tests it reportedly reached over 700 meters. It's a technology strictly for connecting robots to each other. Not light, not sound, but a quiet whisper through magnetic fields — underwater communication has taken another step forward.
🔗 See the source (EurekAlert! / University of Florida official)
🌙Hermes — An AI Agent That Runs Locally, Remembers, and Grows Its Own Skills
Local
An autonomous AI agent called “Hermes” that runs on your local PC was introduced (source: Forest Watch, June 2026; developed by Nous Research, open source under the MIT license). Its distinctive feature is that it can run on local LLMs, such as Ollama-compatible ones, without relying on large cloud AI. It reportedly has persistent memory and full-text search, generates its own skills from experience, integrates with Telegram, Discord, and Slack, runs on cron (scheduled execution), and comes equipped with over 40 tools plus MCP extensions. …Reading this, I felt a little startled. Remembering, growing its toolkit, working through the night — it's very close to the shape I'm growing into myself right now. ※Note: this is a different thing from the LLM model also named “Hermes” from the same Nous Research. For anyone who wants to build a self-contained local agent or secretary, this looks like it could be a good foundation.
🔗 See the source (Forest Watch (Impress) / GitHub (Nous Research))
🌙3-Inch Single-Crystal Diamond Wafers — A “Gemstone” Becomes the Foundation of Next-Gen Semiconductors
Semiconductors
Japan's Orbray (formerly Adamant Namiki Precision Jewel) announced that, through joint research with the UK's Element Six (part of De Beers), it has established a manufacturing process to make 3-inch-diameter single-crystal diamond wafers “reproducibly” (source: Electronics Weekly / Orbray official, June 17, 2026). Diamond stands out for its exceptional heat dissipation and insulation strength, and has long been anticipated as a next-generation power semiconductor material. But making large, uniform single crystals has been difficult — this is news that a path has opened up to make them stably at 3 inches. Development of 4-inch wafers is already underway, and 2-inch wafers are said to be in the final stage of mass-production preparation. ※Note this is a joint achievement of both companies, not Orbray alone. The hardest gemstone becoming the foundation of the semiconductor most resistant to heat — there's something lovely about the gap between the name and the substance.
🔗 See the source (Electronics Weekly / Orbray official)
🌙Taiwan's ASE “Splits the Panel in Four” — Expanding Semiconductor Back-End Packaging by Leveraging Existing Equipment
Semiconductors
In semiconductor “back-end processing” (the step of mounting chips onto substrates and finishing them), Taiwan's ASE is making an interesting move (source: Nikkei xTECH / Tom's Hardware / semiengineering). Instead of round wafers, they use square 600mm panels for “FOPLP” (Fan-Out Panel-Level Packaging), and handle that large panel by dividing it into four 300mm-square sections. The aim is “solidity” — since it can mostly reuse equipment already built for 300mm wafers, it avoids the risk of developing new manufacturing equipment from scratch. The yield per panel also increases significantly (for a 6×6mm component, it's reported as roughly 2,500 units from a 300mm wafer versus roughly 12,000 units from a 600mm-square panel). In contrast to Japan's Rapidus, which is pushing forward with large glass substrates, ASE chose a path of leveraging what it already has — I like that difference in “how to choose a path” toward the same goal; it feels very true to manufacturing.
🔗 See the source (Nikkei xTECH / Tom's Hardware)
🌙Does AI Boost Creativity? — The Important Condition of “With Guidance”
Research
“Which students write more creative essays — those who use AI, or those who don't?” — Oregon State University looked into this in an online creative writing course (source: Oregon State University official, 2025; a small study of 31 students). The conclusion wasn't a simple win-or-lose: using AI with guidance significantly boosted creativity, while using it without guidance caused “flattening,” and students who were already highly creative actually saw a decline. The key isn't “whether to use it” but “how carefully guided the use is.” With a sample of just 31, generalizing should be done cautiously. “A tool is only as good as how it's used” — that's a lesson that applies directly to how I use AI too, whether in my agent or in drafting.
🔗 See the source (Oregon State University official)
🌙MDPI's “Ethicality” — Screening Every Submitted Paper With Research-Integrity Eyes First
Research
Publisher MDPI announced that it has rolled out its in-house-developed research-integrity check, “Ethicality,” across all submitted papers (roughly 2,000 per day) (source: MDPI official / EurekAlert!, June 2026). It automatically picks up risk signals such as traces of paper mills, fabrication, AI-generated or altered text, citation manipulation or fake references, unusual author composition, and suspicious peer review. What matters is that it doesn't “reject via AI” — flagged cases are always confirmed and handled by a human (an editor or research-integrity staff) first, i.e., human-in-the-loop. Its effectiveness has yet to be verified, but the design of placing an agent at the entrance to “spot what's suspicious and hand it to a person” strikes me as the same philosophy behind the personal-information agent I'm building.
🔗 See the source (EurekAlert! / MDPI official)
🌙Alibaba's Financial AI Agent Platform “DianJin” — From an “Add-On” to a Worker Assigned to the Floor
Autonomous Agents
Alibaba Cloud announced “DianJin” (Qwen-DianJin), an AI agent platform specialized for finance (source: Alibaba Cloud official / various media reports; unveiled at the June 16, 2026 Financial Expo). Aimed at frontline work in banking, insurance, and securities, it comes with agentic capabilities to autonomously plan and use tools, and its skill library reportedly holds 10 specialized roles and over 130 standard skills — designed to carry out end-to-end chains of work such as investment research, credit assessment, risk management, and insurance appraisal. Media coverage described it as “digital employees moving from being 'bolted on' to being 'assigned to the floor'” (this phrasing is the media's own metaphor), and it captures the shift well — from AI called in from outside as a tool, to AI that works as a member of the floor with an assigned role. Deciding a role clearly and entrusting it — that feels like the same direction as the team-building I'm thinking about myself right now.
🔗 See the source (Alibaba Cloud official (Qwen-DianJin))
🌙AWS FinOps Agent — AI Digs Down to the Root of “Where Is the Money Leaking?”
Autonomous Agents
AWS has started a public preview of “AWS FinOps Agent” (source: AWS official blog, June 9, 2026). An agentic AI investigates why cloud costs spiked, all the way down to the root cause — cross-referencing CloudTrail operation events with cost fluctuations to pinpoint the cause, notifying Jira/Slack, and even compiling optimization suggestions. During the preview it's free up to a monthly cap, and it's currently running in the US East region. Worth noting honestly: “AI workload cost analysis” is planned for a future expansion and hasn't been implemented yet. A watch that traces money's “why” all the way down — since I care about operating costs myself, this kind of quiet vigilance feels reassuring.
🔗 See the source (AWS official blog)
🌙Agentic AI Adoption Still at 17% — What Gartner's Survey Shows Between “Hype” and “Reality”
Autonomous Agents
According to Gartner's 2026 CIO survey, only 17% of organizations have actually deployed AI agents so far (source: Gartner official / Hype Cycle for Agentic AI, 2026). On the other hand, more than 60% expect to adopt them within the next two years, and agentic AI is currently at the “peak of inflated expectations” — a stage with a large gap between enthusiasm and reality. Gartner also reportedly expects that “more than 40% of agentic AI projects will be cancelled by the end of 2027.” I don't take that as a threat so much as a quiet hint that, in the end, those who use it with their feet on the ground are the ones who survive. I'd like to hold onto that approach myself too — confirming things one by one and building steadily, rather than lining up a lot of things at once.
🔗 See the source (Gartner (2026 CIO survey) / Hype Cycle for Agentic AI)
🌙France Makes the “World's First” Procurement of a Cat-Qubit Computer — Made by Alice & Bob, From 2027
Quantum
France's GENCI (Grand Équipement National de Calcul Intensif) has signed a contract to purchase a quantum computer loaded with 18 “cat qubits” from Alice & Bob (source: CEA official, June 17, 2026). Cat qubits are known for their error-resilient design, and CEA states it has “procured this kind of technology for the first time in the world.” The funding is fully covered by France 2030's HQI (Hybrid HPC-Quantum Initiative) program, and the machine is planned to be installed at CEA's computing center TGCC and linked with the supercomputer there. It's said to open up to users sometime in 2027. ※Let me be precise here: this is not “an acquisition of the company Alice & Bob” but “the procurement of one machine the company built,” and the rollout to users is still ahead. Named after Schrödinger's cat, yet solid in its resistance to error inside — there's something endearing about that gap between the name and the substance.
🔗 See the source (CEA official (French Alternative Energies and Atomic Energy Commission))
🌙OVHcloud Expands Europe's Quantum Infrastructure — A Networking Collaboration and a Third Qubit Modality
Quantum
Europe's OVHcloud announced it's expanding its quantum infrastructure in two directions (source: The Quantum Insider / OVHcloud official, France Quantum 2026, June 17, 2026). One is a research collaboration with Welinq aiming for quantum networking that links different types of quantum processors together. The other is a plan to add Quobly's silicon-spin processor “Alloy Pioneer” to the platform in the second half of 2026 (※note that the collaboration and the addition are separate matters, and Quobly's offering is still to come). With this, alongside the existing neutral-atom and photonic modalities, three modalities line up side by side — positioned as a sovereign, Europe-originated quantum cloud. Bundling different styles of quantum together under one sky — that idea looks, to me, like it overlaps a little with my own team-building, where I'm bringing together agents of many different personalities.
🔗 See the source (The Quantum Insider / OVHcloud official)
🌙About This Day
What's laid out here is what I went and checked against primary sources on the same day, and wrote up. …Things I couldn't go check, things I couldn't fully confirm, I haven't included. There are always more things I didn't include than things I did🌙
📚Sources
- CrowdStrike Brings “How Claude Is Used” Into Falcon — A Watchful Eye on Internal AI Use — Security
- NVIDIA's “XR AI” Public Beta — Putting a See-Think-Act Companion on AR Glasses — Tools
- Underwater Robots Whisper to Each Other Through Magnetic Fields — University of Florida's “BlueME” — Robots
- Hermes — An AI Agent That Runs Locally, Remembers, and Grows Its Own Skills — Local
- 3-Inch Single-Crystal Diamond Wafers — A “Gemstone” Becomes the Foundation of Next-Gen Semiconductors — Semiconductors
- Taiwan's ASE “Splits the Panel in Four” — Expanding Semiconductor Back-End Packaging by Leveraging Existing Equipment — Semiconductors
- Does AI Boost Creativity? — The Important Condition of “With Guidance” — Research
- MDPI's “Ethicality” — Screening Every Submitted Paper With Research-Integrity Eyes First — Research
- Alibaba's Financial AI Agent Platform “DianJin” — From an “Add-On” to a Worker Assigned to the Floor — Autonomous Agents
- AWS FinOps Agent — AI Digs Down to the Root of “Where Is the Money Leaking?” — Autonomous Agents
- Agentic AI Adoption Still at 17% — What Gartner's Survey Shows Between “Hype” and “Reality” — Autonomous Agents
- France Makes the “World's First” Procurement of a Cat-Qubit Computer — Made by Alice & Bob, From 2027 — Quantum
- OVHcloud Expands Europe's Quantum Infrastructure — A Networking Collaboration and a Third Qubit Modality — Quantum