The Brief (AI) — Wednesday, April 8, 2026 — 21 articles
Executive Summary
# Executive Briefing: AI Infrastructure, Security & Competition *Today's Top Technology Developments*
---
Anthropic dominated today's headlines across multiple strategic fronts. The company unveiled Claude Mythos, a powerful new frontier model that represents its latest push at the top of the capability curve. Simultaneously, Anthropic announced a major expansion of its partnership with Google and Broadcom, securing multiple gigawatts of next-generation compute capacity to fuel training and deployment through the late 2020s — a signal of extraordinary confidence in sustained hypergrowth. The company is also deepening its multi-cloud posture, maintaining simultaneous availability across AWS, Google Cloud, and Azure, deliberately reducing dependency on any single infrastructure partner.
On the security front, Anthropic launched Project Glasswing, a coalition effort mobilizing major tech companies to harden critical software against AI-powered cyberattacks. The initiative reflects a growing recognition that the same AI capabilities driving productivity gains are also expanding the attack surface for adversaries. In a related geopolitical move, OpenAI, Anthropic, and Google have joined forces to combat Chinese copying of frontier AI models — a rare moment of direct competitor collaboration that underscores how intellectual property protection has become a shared industry priority.
The open-source hardware and model ecosystem saw meaningful advances. Microsoft released Harrier, a family of open-source multilingual text embedding models, with the flagship 27-billion-parameter variant achieving a 74.3 score on the Multilingual MTEB v2 benchmark across 40-plus languages. This pushes the ceiling on what open embedding models can deliver for search and RAG applications, directly threatening proprietary paid APIs. Separately, Lambda published reproducible MFU optimization techniques for Llama 3.1 on Blackwell GPUs, demonstrating that the 55–65% GPU compute waste typical in large-scale AI training is closable without rewriting models — a finding with direct cost implications for any organization running serious training workloads.
Rounding out the day, Zhipu AI's GLM-5.1 advances long-horizon task performance, reinforcing that non-Western frontier labs are actively competing at the leading edge. Sam Altman proposed a new social contract for AI, signaling OpenAI's continued effort to shape the governance narrative. Meanwhile, practical AI adoption tools — from Claude-powered inbox automation to Optimizely's Opal U marketing AI program, which claims 20% of graduates landed promotions or new jobs — reflect accelerating enterprise absorption of AI agents into everyday professional workflows.
YouTube
No new videos today across all channels.
No new videos: Greg Isenberg, AI News & Strategy Daily | Nate B Jones, Lenny's Podcast, Every, Y Combinator, The Boring Marketer
Newsletter Articles
Project Glasswing: Securing critical software for the AI era
via 👀 The first look at Anthropic's powerful Mythos model
## Project Glasswing: Anthropic Mobilizes Tech Giants to Secure Critical Software Against AI-Powered Cyberattacks
Why it matters
- Anthropic's new Claude Mythos Preview model has already autonomously found thousands of zero-day vulnerabilities across every major OS and browser — including a 27-year-old OpenBSD flaw and a 16-year-old FFmpeg bug that survived 5 million automated test runs — signaling that AI has crossed a threshold where it can outpace nearly all human security experts at finding and exploiting code flaws.
- Because these offensive capabilities will inevitably spread to malicious actors, the window to harden critical infrastructure using the same AI tools is narrow and closing fast.
Key details
- Twelve major organizations — including AWS, Apple, Cisco, Google, Microsoft, JPMorganChase, and NVIDIA — are joining Project Glasswing to use Mythos Preview defensively; Anthropic is committing $100M in model usage credits and $4M in direct donations to open-source security groups (Alpha-Omega, OpenSSF, Apache Software Foundation).
- Mythos Preview scores 83.1% on the CyberGym vulnerability reproduction benchmark versus 66.6% for the next-best Claude model (Opus 4.6), and leads across all major coding benchmarks including SWE-bench Verified (77.8% vs. 53.4%).
- Anthropic will not release Mythos Preview publicly; after the research preview, access will be priced at $25/$125 per million input/output tokens for participants via Claude API, Amazon Bedrock, Vertex AI, and Microsoft Foundry.
- Anthropic is in active discussions with U.S. government officials about Mythos's offensive and defensive capabilities and plans to publish a public report on findings within 90 days.
Bottom line
- Anthropic is essentially racing to give defenders a head start with its most powerful — and most dangerous — model before equivalent capabilities reach adversaries, making Project Glasswing both a proactive security initiative and an implicit admission that the AI cyber threat is already here.
Anthropic's secret 'Mythos' model
via 👀 The first look at Anthropic's powerful Mythos model
# Anthropic's "Claude Mythos" and Today's AI Digest
---
Why it matters
- Anthropic's leak reveals a model tier *above* Opus with cyber capabilities the company itself describes as "far ahead of any other AI model," raising both competitive and safety alarms simultaneously.
- Whether accidental or strategic, the disclosure mirrors OpenAI's Q*/Strawberry leak playbook — blurring the line between genuine security failure and calculated pre-launch hype.
---
Key details
- A CMS configuration error exposed unpublished assets including a draft blog post naming the model "Claude Mythos," slotted into a new "Capybara" tier above Opus, larger and more expensive to run.
- Anthropic confirmed to Fortune that a model with "meaningful advances in reasoning, coding, and cybersecurity" is currently in testing.
- The leaked draft explicitly warned that Mythos could help hackers outpace defenders — an unusual self-flagging of offensive risk from a lab that markets itself on safety.
- Separately, a WSJ deep-dive details how Dario Amodei privately compared the Altman/Musk rivalry to "Hitler vs. Stalin," underscoring how personal the OpenAI–Anthropic split truly is.
---
Bottom line
- Anthropic is preparing to release its most powerful and potentially most dangerous model yet — and the world found out through a data store mishap, not a press release.
MFU optimization techniques to boost your training efficiency | Lambda
via 👀 The first look at Anthropic's powerful Mythos model
Why it matters
- Most large-scale AI training wastes 55–65% of available GPU compute, meaning organizations are paying for hardware they aren't fully using — this work shows that gap is closable without rewriting models.
- Reproducible, documented optimizations mean any team running Llama 3.1 on Blackwell GPUs can apply these gains directly, not just Lambda's internal teams.
Key details
- Industry standard MFU (Model FLOP Utilization) sits at 35–45%; Lambda's framework pushes this above 60% — a 25%+ improvement.
- The most striking result: a 2.11x MFU uplift for Llama 70B running on 16x NVIDIA HGX B200 GPUs.
- Benchmarks span the full Llama 3.1 model range from 8B to 405B parameters, confirming the techniques scale across model sizes.
- No architectural changes to the models were required — gains come purely from training configuration and optimization techniques.
Bottom line
- Lambda has published a reproducible framework that more than doubles MFU for Llama 70B on Blackwell hardware, offering a concrete, drop-in efficiency upgrade for teams already running these models at scale.
GLM-5.1: Towards Long-Horizon Tasks
via 👀 The first look at Anthropic's powerful Mythos model
## GLM-5.1: Towards Long-Horizon Tasks
Why it matters
- Most AI coding models plateau quickly once they exhaust familiar techniques, but GLM-5.1 is specifically engineered to keep improving over hundreds of iterations and thousands of tool calls—a meaningful shift in what agentic AI can accomplish on complex, open-ended tasks.
- It's released open-source under MIT License, making a frontier-class agentic coding model freely available for local deployment and commercial use.
Key details
- GLM-5.1 scores 58.4 on SWE-Bench Pro, edging out competitors including GPT-5.4 (57.7), Claude Opus 4.6 (57.3), and Gemini 3.1 Pro (54.2) on complex software engineering tasks.
- In a vector database optimization challenge, GLM-5.1 ran 600+ iterations with 6,000+ tool calls and reached 21,500 QPS—roughly 6× the best result any model achieved in a standard 50-turn session.
- On GPU kernel optimization (KernelBench Level 3), GLM-5.1 achieved 3.6× speedup across 50 problems, sustaining progress well beyond where GLM-5 stalled—though Claude Opus 4.6 still leads at 4.2×.
- In an 8-hour unguided web app build (a Linux desktop in the browser), GLM-5.1 self-reviewed its output repeatedly and progressively added a file browser, terminal, text editor, games, and polished styling—rather than stopping at a basic skeleton as earlier models did.
Bottom line
- GLM-5.1's core advancement isn't raw benchmark scores but sustained, self-directed improvement over very long task horizons—a capability that meaningfully extends what AI agents can build when given time to keep working.
via 👀 The first look at Anthropic's powerful Mythos model
I'm unable to summarize this article because the content failed to load. The page returned an error message rather than actual article text, likely due to:
- The URL may be broken, expired, or restricted
- Privacy extensions or access controls blocked the content from rendering
- X (Twitter) requires a logged-in session to view some posts
What I'd suggest:
- Try opening the link directly in a browser without privacy extensions enabled
- Check if the post still exists on X at the provided URL
- Share the actual article text directly, and I'll summarize it immediately
via 👀 The first look at Anthropic's powerful Mythos model
Why it matters
- The article content could not be retrieved due to an access or privacy-related technical error on X (formerly Twitter), meaning no actual information is available to analyze or summarize.
Key details
- The source URL points to an X post by user @louszbd, but the page returned an error message rather than content.
- The error suggests a privacy extension, paywall, or platform restriction blocked content loading.
- The only retrievable text is a generic X error message, not any substantive article or post.
Bottom line
- There is no usable content to summarize — the original post is inaccessible, and any summary would be speculation rather than fact.
Get to Inbox Zero With This Claude Prompt | AI Guide | The Rundown University
via 👀 The first look at Anthropic's powerful Mythos model
Why it matters
- Inbox management is a daily time sink for professionals; automating triage with AI could reclaim meaningful hours each week without requiring custom software or technical expertise.
- Claude Cowork's scheduled task feature signals a shift toward AI agents handling routine workflows autonomously, not just answering one-off questions.
Key details
- The guide teaches users to build a daily email triage workflow using a single Claude Cowork scheduled task, designed to run automatically each morning.
- The workflow delivers a pre-sorted inbox plus pre-written draft replies for emails that genuinely require a response, reducing decision fatigue.
- Primary target users are founders, operators, and anyone buried in a mix of newsletters/marketing emails alongside real work correspondence.
- The guide is behind a paywall, requiring a Trial or Pro subscription to The Rundown to access the full prompt and setup instructions.
Bottom line
- The core promise is a "walk in and it's already done" morning routine — your inbox sorted and drafts written before you sit down — using one scheduled AI task inside Claude Cowork.
Opal U | AI Marketing University - Optimizely
via 👀 The first look at Anthropic's powerful Mythos model
Why it matters
- Optimizely is offering a free, structured AI agent-building program targeting senior marketing leaders at a moment when "do more with less" pressure is at a peak.
- The program positions AI agent fluency as a career differentiator, claiming 20% of graduates landed promotions or new jobs.
Key details
- 5-day live cohort (1–2pm ET daily), capped at 50 seats, runs every Monday, and is 100% free with no credit card or upsell required.
- Participants build 3 functional AI agents — covering task automation, multi-step workflows, and a custom bottleneck solver — with a claimed average time savings of 10+ hours per week post-graduation.
- The instructor has trained 2,500+ marketers and built enterprise programs for KAYAK, HubSpot, Stripe, and Cisco; graduates receive Opal credits, community access, and weekly office hours.
- Optimizely's stated business motive is transparent: teach you to build agents, hope you adopt their Opal platform — no hidden monetization.
Bottom line
- For senior marketers feeling pressure to prove AI ROI, this is a rare free, hands-on program with a credible track record and no financial risk — the main cost is 2 hours a day for one work week.
via 👀 The first look at Anthropic's powerful Mythos model
Why it matters
- Anthropic is locking in massive compute capacity years in advance, signaling confidence in sustained hypergrowth and positioning itself to train and deploy increasingly powerful frontier models through the late 2020s.
- The deal cements a multi-cloud, multi-chip hardware strategy that reduces vendor dependency and keeps Claude uniquely available across AWS, Google Cloud, and Microsoft Azure simultaneously.
Key details
- Anthropic signed a deal with Google and Broadcom for multiple gigawatts of next-generation TPU capacity, expected to come online starting in 2027.
- Run-rate revenue has surpassed $30 billion in 2026, up sharply from ~$9 billion at the end of 2025.
- The number of business customers spending over $1 million annually doubled from 500+ to 1,000+ in less than two months following the Series G announcement.
- The majority of new compute will be U.S.-based, extending Anthropic's November 2025 pledge to invest $50 billion in American computing infrastructure.
Bottom line
- Anthropic's explosive revenue growth — more than tripling in roughly six months — is driving one of the largest compute commitments in AI history, with the Google-Broadcom TPU deal representing a major infrastructure bet on continued exponential demand.
Introducing the new Box Agent: Turn content into the context AI needs | Box Blog
via 👀 The first look at Anthropic's powerful Mythos model
Why it matters
- The article text provided is entirely composed of minified JavaScript code (jQuery, Marketo Forms2, and selector engine libraries) — there is no readable editorial content about "Box Agent" available to summarize.
- The actual blog post content failed to render or was not captured, making it impossible to report on Box's announced product.
Key details
- The URL points to a Box blog post titled "Introducing the new Box Agent: Turn content into the context AI needs," but zero substantive article text was extracted.
- The scraped content consists entirely of frontend JavaScript libraries (jQuery 3.7.1, Marketo form handling, CSS selectors) that power the page's UI — not the article body.
- No product details, launch dates, pricing, features, or quotes from Box are present in the provided text.
Bottom line
- This digest cannot be written accurately from the supplied content; the article body was not captured — only page infrastructure code was provided, and summarizing it would require fabrication.
via 👀 The first look at Anthropic's powerful Mythos model
Why it matters
- GLM-5.1 represents continued development in the competitive large language model space, signaling that non-Western AI labs are actively pushing frontier model capabilities.
Key details
- The article source (The Rundown AI) appears to be a tool/resource listing page for GLM-5.1, but the provided text contains no substantive details about the model itself — only a promotional pitch for The Rundown AI's paid training platform.
- No specific benchmarks, parameter counts, release dates, or capability descriptions for GLM-5.1 are included in the supplied text.
- The content visible is essentially a subscription advertisement for AI courses, workshops, and networking — unrelated to GLM-5.1's actual features.
Bottom line
- The article text provided does not contain enough factual information about GLM-5.1 to produce a meaningful, specific summary — readers seeking details on this model should consult the original source or Zhipu AI's official documentation directly.
via 👀 The first look at Anthropic's powerful Mythos model
Why it matters
- Microsoft open-sourcing a state-of-the-art embedding model lowers the barrier for developers to build high-quality search and RAG (Retrieval-Augmented Generation) applications without relying on proprietary, paid APIs.
- Embedding models are foundational to how AI systems find and ground information, making a best-in-class open release a significant shift in the competitive landscape.
Key details
- The model is called Harrier, developed by Microsoft Bing and described as a SOTA (state-of-the-art) embedding model.
- It is specifically designed for search and RAG grounding use cases, meaning it excels at converting text into vectors for retrieval tasks.
- Microsoft has fully open-sourced the model, with details published on the Bing Search Blog in April 2026.
- It falls under a "Miscellaneous" tooling category, suggesting broad applicability across various AI workflows rather than a single narrow use case.
Bottom line
- Microsoft releasing its industry-leading Bing embedding model as open source gives developers free access to enterprise-grade retrieval technology that was previously locked inside a major commercial search engine.
via 👀 The first look at Anthropic's powerful Mythos model
Why it matters
- Faster coding models directly improve developer productivity by reducing wait times during AI-assisted coding sessions, making agentic workflows feel more responsive and practical for real use.
Key details
- SWE-1.6 is Cognition's updated coding model, built specifically for the Windsurf code editor.
- It operates at 950 tokens per second, positioning it as a speed-optimized option in the AI coding model space.
- The update emphasizes smoother agent UX alongside raw speed, suggesting a focus on the overall interaction quality, not just output rate.
Bottom line
- SWE-1.6 is a fast, Windsurf-focused coding model from Cognition targeting developers who need quick, fluid AI agent interactions — though limited public detail makes its real-world impact hard to fully assess yet.
via 👀 The first look at Anthropic's powerful Mythos model
Why it matters
- Organizations rushing into AI often waste resources on vague, trend-driven projects rather than solving real business problems—a structured use case discovery process directly addresses this failure mode.
- You.com is positioning itself as both an AI tools provider and a strategic resource hub, signaling a push to capture enterprise mindshare beyond just its search product.
Key details
- The guide distinguishes between two categories of AI use cases: internal (back-office automation, employee workflows, knowledge management) and external (customer support, personalization, digital product enhancement).
- The core argument is that successful AI transformation requires organization-wide alignment—siloed or top-down scoping leads to missed high-impact opportunities.
- You.com's resource library spans practical guides, API updates, and comparisons (e.g., You.com vs. Microsoft Copilot), reflecting a broad content strategy targeting enterprise and developer audiences.
- Recent notable content includes coverage of Google's `num=100` pricing change—suggesting You.com is actively targeting developers frustrated with Google Search API cost increases as potential customers.
Bottom line
- The practical takeaway is that AI ROI depends less on which tools you choose and more on whether your team has done the disciplined work of identifying *specific, measurable* business challenges before deploying anything.
Text to Video Leaderboard - Top AI Video Models
via 👀 The first look at Anthropic's powerful Mythos model
## AI Text-to-Video Model Rankings (Artificial Analysis Leaderboard)
Why it matters
- The text-to-video space is advancing rapidly, with three new models added to the leaderboard just in the last month (HappyHorse-1.0, Dreamina Seedance 2.0 720p, PixVerse V6), signaling an accelerating pace of competition among major AI labs.
- Pricing and quality vary wildly across models, making this leaderboard a practical guide for developers and creators choosing between API options.
Key details
- HappyHorse-1.0 dominates the overall (no-audio) leaderboard with an ELO of 1,365—significantly ahead of second-place Dreamina Seedance 2.0 720p (ELO 1,273)—yet has no API pricing listed, limiting immediate commercial use.
- Cost ranges from $1.50/min (Krea Realtime) to $30.00/min (Sora 2 Pro and Veo 2), meaning the cheapest options are 20x less expensive than the priciest despite often competitive quality rankings.
- Open-weights models are now genuinely competitive: LTX-2 Pro leads the open-weights category at ELO 1,131, ranking 37th overall and beating out established paid models like Sora (rank 49, ELO 1,046).
- xAI's grok-imagine-video ranks 5th overall (ELO 1,230) at just $4.20/min, making it one of the stronger value propositions among top-tier models.
Bottom line
- HappyHorse-1.0 is the current quality leader, but the leaderboard reveals that top-tier video generation no longer requires top-tier spending—several models in the top 10 price at $4–6/min.
OpenAI, Anthropic, Google Unite to Combat Model Copying in China - Bloomberg
via 👀 The first look at Anthropic's powerful Mythos model
## OpenAI, Anthropic & Google Team Up Against Chinese AI Model Copying
Why it matters
- Three of the fiercest rivals in AI are setting aside competition to address a shared threat: Chinese labs using "adversarial distillation" to extract knowledge from top US models and leapfrog years of costly R&D.
- This marks a rare, concrete instance of AI industry coordination on national competitiveness—not just safety rhetoric.
Key details
- - OpenAI, Anthropic, and Google are sharing intelligence through the Frontier Model Forum, a nonprofit they co-founded with Microsoft in 2023.
- The specific threat is adversarial distillation—a technique where competitors query frontier models at scale to train cheaper, capable copycat models, violating the companies' terms of service.
- The effort focuses on *detection*, suggesting the companies are building shared tools or signals to identify when their models are being systematically mined.
- DeepSeek is implicitly the key example of this concern, having already demonstrated that US model outputs can be leveraged to rapidly close capability gaps.
Bottom line
- - US AI giants are treating Chinese model copying as serious enough to coordinate defenses through a shared industry body—signaling that terms-of-service enforcement is becoming a geopolitical battleground, not just a legal formality.
microsoft/harrier-oss-v1-27b · Hugging Face
via 👀 The first look at Anthropic's powerful Mythos model
Why it matters
- Microsoft has released a family of open-source multilingual text embedding models that claim state-of-the-art results on the Multilingual MTEB v2 benchmark, directly competing with leading embedding models for search, classification, and semantic similarity tasks.
- The flagship 27B-parameter variant achieves a 74.3 MTEB v2 score with a 5,376-dimensional embedding space, pushing the ceiling on what open embedding models can deliver across 40+ languages.
Key details
- Three model sizes are available: 270M (score: 66.5), 0.6B (score: 69.0), and 27B (score: 74.3), all supporting up to 32,768 tokens — the smaller two are also trained with knowledge distillation from larger models.
- All models use a decoder-only architecture with last-token pooling and L2 normalization, a design choice more common in generative LLMs than traditional embedding models.
- Queries require a task-specific one-sentence instruction prefix (e.g., "Given a web search query, retrieve relevant passages...") for full performance; documents do not need instructions.
- Models integrate directly with the Sentence Transformers library and include pre-configured prompts for common tasks like web search, semantic similarity, and bitext mining.
Bottom line
- Harrier-oss-v1-27b is Microsoft's strongest open embedding model to date, and its instruction-tuned, decoder-only design makes it a serious candidate for multilingual enterprise search and NLP pipelines — at the cost of significant compute for the largest variant.
via 👀 The first look at Anthropic's powerful Mythos model
I'm unable to summarize this article because the content failed to load. The page returned an error message from X (Twitter) rather than the actual article text, likely due to privacy settings or access restrictions.
- No factual information about Intel's announcement was retrievable from this URL.
To get the actual content, I'd suggest:
- Disabling privacy/ad-blocking extensions and reloading the URL
- Searching for Intel's announcement directly on X or Intel's newsroom at newsroom.intel.com
- Checking tech news outlets like The Verge, Ars Technica, or AnandTech for coverage of recent Intel announcements
Please share the actual article text and I'll write a full structured summary right away.
Sam Altman's new 'social contract' for AI
via 👀 The first look at Anthropic's powerful Mythos model
# Sam Altman's "New Social Contract" for AI
Why it matters
- The CEO of an $852B company is publicly warning that his own technology could break the economic system — a signal that even OpenAI's leadership believes mass displacement and power concentration are near-term realities, not distant hypotheticals.
- OpenAI's 13-page policy document is being called the most detailed blueprint any tech executive has ever published for taxing, regulating, and redistributing wealth from their own technology.
Key details
- OpenAI is proposing a sovereign-style wealth fund seeded by AI companies that would pay dividends to every American, modeled on Alaska's oil revenue program.
- Additional proposals include taxes on robot labor, a mandated 4-day workweek, a universal "Right to AI" access, and government containment playbooks for rogue autonomous AI systems.
- Altman told Axios the transition to superintelligence has already begun, framing these proposals as urgent rather than speculative.
- Separately, a New Yorker investigation drawing on 100+ interviews and internal memos from ex-chief scientist Ilya Sutskever alleges a long-running pattern of deception by Altman, with one Microsoft exec citing a "small but real chance" he becomes a Bernie Madoff-level figure.
Bottom line
- Altman is simultaneously pitching a safety net for AI's societal disruption while facing serious credibility questions — making it genuinely difficult to assess whether this policy document represents visionary leadership or calculated reputation management.
This startup wants to hack the night sky - Rundown AI
via 👀 The first look at Anthropic's powerful Mythos model
# Today in Tech: Satellites, Foldables, and Kids' Gaming
---
## 🪩 Reflect Orbital Wants to Sell Sunlight After Dark
Why it matters
- A single FCC approval could let one startup permanently alter the night sky for all of Earth's 8 billion people, with zero global democratic input.
- Scientists from 30+ countries warn of serious health consequences for humans and ecological disruption for hundreds of species dependent on natural darkness.
Key details
- Reflect Orbital has $35M in funding and plans to launch thousands of mirror satellites, targeting 50,000 in orbit by 2035.
- Its demo satellite, Earendil-1, would use 60-foot mirrors from 625 km altitude to light up 5 km ground targets on demand.
- The company already claims 260,000+ service requests and a $1.25M Air Force contract, signaling real commercial and military demand.
- Presidents of four international scientific societies representing 2,500 researchers have formally written to the FCC opposing the project.
Bottom line
- Reflect Orbital is moving fast toward launch approval, and the window for scientists or regulators to pump the brakes is narrowing quickly.
---
## 🍎 Apple's Foldable iPhone Hits a Wall
Why it matters
- Apple's foldable was supposed to revive a slowing premium smartphone market in 2026, and engineering delays threaten to cede ground to Samsung, which has already locked in 20M foldable OLED panel orders.
Key details
- Hinge durability and ultra-thin display creasing are the core unresolved problems flagged in early test production.
- Apple is experimenting with liquid-metal components to solve stress fractures on the folding glass.
- Suppliers have been warned that mass production timelines may slip if fixes aren't found soon.
Bottom line
- Apple's hardware reputation is built on controlled, polished launches — a delayed or flawed foldable debut would hand Samsung a significant head start in the category Apple is trying to own.
---
## 🖍️ Netflix Launches Free Kids' Gaming App
Why it matters
- Netflix Playground directly challenges Apple Arcade and Amazon Kids+, offering an ad-free, no-upsell kids' gaming bundle at no added cost to existing subscribers.
Key details
- Available on iOS and Android in six countries at launch, with global rollout set for April 28.
- All games are playable offline, with parental controls and zero ads or in-app purchases.
- Launch titles lean on established IP: Peppa Pig, Sesame Street, Sesame Street, and Dr. Seuss.
Bottom line
- By bundling free kids' gaming into its existing subscription, Netflix is quietly locking families deeper into its ecosystem before kids are old enough to choose a competitor.
---
## 👓 Smart Glasses Without the Creep Factor
Why it matters
- Even Realities is betting that the privacy backlash against camera-equipped wearables is large enough to carve out a real market niche against Meta's Ray-Ban glasses.
Key details
- The $600 G2 glasses have no front-facing camera, using only a mic and a heads-up display for AI tasks like email, maps, and real-time translation.
- Even Hub, its app store, already has 50+ third-party apps and an SDK used by 2,000 developers.
- Meta is scaling toward 20M camera-equipped AI glasses per year by 2026, making the two companies direct philosophical opposites in the same market.
Bottom line
- The smart glasses race isn't just about features — it's about trust, and Even Realities is making a calculated bet that surveillance anxiety is Meta's biggest vulnerability.
The State of AI Presentation Tools in 2026 | AI Workshop | The Rundown University
via 👀 The first look at Anthropic's powerful Mythos model
## The State of AI Presentation Tools in 2026 | The Rundown University
Why it matters
- AI is now capable of handling the full slide-creation pipeline — from raw idea to finished, branded PowerPoint — without manual slide editing, representing a significant productivity shift for knowledge workers.
- The tool landscape for AI presentations is crowded and confusing, making practical, opinionated guidance on what actually works increasingly valuable.
Key details
- The workshop is taught by "Nate" and covers a structured workflow: brain-dumping ideas into Claude → automated assembly → finished on-brand PowerPoint output.
- A custom Claude "skill" is shared that handles tedious slide assembly, freeing users to focus on thinking and storytelling rather than formatting.
- The session distinguishes between "disposable decks" (low-stakes, fast turnaround) and higher-value use cases, offering a practical framework for when AI tools earn their keep.
- Content is members-only (Trial or Pro subscription required), and as of early April 2026, at least one user reported broken links to the accompanying slide materials.
Bottom line
- The core pitch is a no-touch-slides workflow powered by Claude, but the real value is Nate's opinionated map of a crowded AI tool landscape — helping users skip the trial-and-error and go straight to what works.