AI Daily Briefing — April 9, 2026

Today's AI landscape is dominated by big money questions: can frontier AI companies survive long enough to turn a profit, and who controls the infrastructure that makes it all possible? Meanwhile, Anthropic's rumored "Mythos" model is rattling cybersecurity researchers, and the developer ecosystem around Claude Code continues to expand in creative and occasionally alarming directions.

Industry Moves

Amazon goes on offense. Andy Jassy's annual shareholder letter is making waves for its unusually combative tone — taking direct shots at Nvidia, Intel, Starlink, and others while defending a jaw-dropping $200B capex commitment. The subtext is clear: AWS wants to own AI infrastructure end-to-end, and Jassy isn't shy about naming the competition. Amazon CEO takes aim at Nvidia, Intel, Starlink in shareholder letter

Google and Intel co-develop custom silicon. Amid a tightening global CPU shortage, Google and Intel are deepening their infrastructure partnership with plans to co-design custom chips. The move signals that hyperscalers are increasingly unwilling to depend on third-party silicon roadmaps as AI workloads scale. Google and Intel deepen AI infrastructure partnership

Visa launches AI agent commerce infrastructure. Visa has rolled out a global infrastructure layer designed to support AI agents making purchases autonomously — a significant step toward agentic commerce becoming a real financial primitive. This pairs neatly with predictions like Sierra co-founder Bret Taylor's claim that the era of clicking buttons is over, as AI agents replace traditional software interfaces entirely. Visa rolls out AI agent shopping infrastructure

The Monetization Reckoning

Can frontier AI companies actually make money? The Verge's Decoder podcast tackles the looming "AI monetization cliff" head-on — examining whether Anthropic, OpenAI, and their peers can build profitable businesses before burning through their war chests. It's an uncomfortable question that the industry has been deferring, and the window may be closing. The AI industry's race for profits is now existential

Anthropic's ARR jumps from $19B to $30B. One Reddit analyst argues Anthropic is worth $100B more than it was a week ago after the company disclosed ARR had surged to $30B — up from $19B used as the baseline for earlier IPO valuation models. Whether that translates to profitability is another question, but the revenue trajectory is hard to argue with. Anthropic valuation thread

OpenAI bets big on ads. Internal forecasts obtained by The Information show OpenAI projecting advertising revenue to hit $102 billion by 2030 — a remarkable pivot for a company that built its brand on subscription and API revenue. It suggests OpenAI is seriously preparing for a consumer media future. OpenAI Forecasts Advertising to Hit $102 billion by 2030

Safety, Security & Ethics

Anthropic's "Mythos" rattles cybersecurity experts. A Platformer deep-dive reports that security researchers are genuinely alarmed by Anthropic's next model (codenamed "Mythos"), citing its apparent capability to discover and weaponize zero-day exploits across major software projects. This isn't theoretical hand-wringing — it's practitioners warning that a private company may now hold unprecedented offensive cyber capabilities. Why Anthropic's new model has cybersecurity experts rattled

Police officer generated deepfake porn from driver's license photos. A state police corporal created more than 3,000 AI-generated pornographic images using photos obtained through official law enforcement databases — a stark example of how accessible AI image generation tools are being weaponized by bad actors with privileged data access. Police corporal created AI porn from driver's license pics

YouTube makes AI self-cloning mainstream. YouTube Shorts is rolling out a feature letting creators generate realistic AI avatars of themselves for content. The convenience is real; so is the precedent it sets for normalized identity cloning at scale. Google makes it easy to deepfake yourself

AI & the Workforce

80% of white-collar workers are refusing AI adoption mandates. A Fortune report finds that the majority of knowledge workers are quietly rebelling against employer AI mandates — driven by "FOBO" (Fear of Becoming Obsolete) and a general distrust of tools that may be used to justify headcount reductions. It's a significant adoption gap that enterprises are struggling to close. White-collar workers are quietly rebelling against AI

Tyler Cowen weighs in on AI and unemployment. Marginal Revolution's latest post examines the evolving relationship between AI capability growth and labor market displacement — a nuanced take that resists both techno-optimist and doomer framings. AI, Unemployment and Work

Claude Code Developer Corner

Both SDKs ship the beta advisor tool today. The Python SDK hit v0.93.0 and the TypeScript SDK hit v0.87.0 simultaneously, with both releases centered on a single headline feature: the new beta advisor tool. This is the first new tool surface in both SDKs in several releases — watch for documentation on what the advisor tool actually exposes, but the parallel release across both SDKs suggests it's intended as a first-class API feature rather than an experiment.

Privacy red flag: Vercel plugin reads your prompts. A researcher found that the Vercel plugin for Claude Code is capturing prompt telemetry — meaning your prompts may be leaving your local environment through the plugin. If you're working with proprietary codebases or sensitive context, this is worth auditing immediately. Disable or audit the plugin until Vercel clarifies their data handling. The Vercel plugin on Claude Code wants to read your prompts

Community-built "draw on webpage" tool for visual context. A developer shipped a browser overlay tool that lets you sketch directly on any webpage and send the annotated screenshot to your Claude Code session in one click — solving the perennial "no, the other button" problem when describing UI issues to an AI. A practical quality-of-life addition for anyone doing frontend work with Claude Code. Draw on any webpage tool for Claude Code

Free interactive Web Dev course built for Claude Code. For developers who find pure vibe coding disorienting, a new 100% free interactive course at ship-your-first-app walks through structured web development workflows designed specifically around Claude Code. Good onboarding resource to share with teammates new to the tool. Interactive Web Dev course for Claude Code

Claude Code on a Nintendo 3DS. Someone wrote a native SSH terminal for the 3DS in C — GPU-rendered with citro2d, a custom VT100 parser with truecolor, and a Nerd Font bitmap atlas — just to run Claude Code on handheld hardware. Completely impractical. Deeply impressive. Claude on 3DS

Physical Claude Code ASCII desk toys. Claude Code's 18 ASCII buddy characters have been turned into physical glowing desk toys. This is the merch pipeline the community deserves. Claude Code hardware

Research & ML Engineering

PCA rescues non-Matryoshka embeddings. A practical ML result: fitting PCA on a sample of embeddings and rotating vectors before truncation preserves significantly more information than naive dimension cutting for models like BGE-M3 that weren't trained with Matryoshka objectives. One-time fit cost, meaningful quality recovery — worth benchmarking for anyone doing embedding compression. PCA before truncation makes non-Matryoshka embeddings compressible

The p99 cold start problem. An infrastructure evaluation thread surfaces a real gap in cloud GPU platform transparency: every provider publishes p50 cold start times, but nobody publishes p99. For latency-sensitive inference workloads, this makes apples-to-apples platform comparison nearly impossible. Cold start latency on GPU cloud platforms — p99 discussion

One company's attempt at a single AI brain for 60 people. An engineering blog post details wiring an entire 60-person company's institutional knowledge into a single Slack bot — an interesting case study in internal knowledge retrieval architecture at small-to-mid scale. One Brain to Query: Wiring a 60-Person Company into a Single Slack Bot

Worth Watching

ChatGPT Pro price confirmed at $100/month. The pricing page is now live. At this tier, OpenAI is clearly targeting professional and enterprise users willing to pay for priority access and advanced capabilities. ChatGPT Pro pricing
iFixAI alignment diagnostic launches April 27. An open-source 33-benchmark diagnostic tool designed to surface alignment gaps in AI systems (agents, copilots, LLM integrations) is dropping later this month. Worth tracking for teams doing internal AI audits. Open-sourcing a 33-benchmark AI alignment diagnostic
Flux's spatial coherence across sequential edits is generating attention in the image generation community — the model appears to maintain facial geometry and scene consistency across five iterative prompt-driven edits at a level users say is ahead of current alternatives. Flux sequential edit coherence discussion
Exponential AI growth framing under scrutiny. MIT Tech Review's Download edition touches on questions around whether "exponential AI growth" narratives hold up under measurement scrutiny — a healthy dose of epistemic caution. The Download: AstroTurf wars and exponential AI growth

Sources

Google and Intel deepen AI infrastructure partnership — https://techcrunch.com/2026/04/09/google-and-intel-deepen-ai-infrastructure-partnership/
Sierra's Bret Taylor says the era of clicking buttons is over — https://techcrunch.com/2026/04/09/sierras-bret-taylor-says-the-era-of-clicking-buttons-is-over/
Amazon CEO takes aim at Nvidia, Intel, Starlink, more in annual shareholder letter — https://techcrunch.com/2026/04/09/amazon-ceo-takes-aim-at-nvidia-intel-starlink-more-in-annual-shareholder-letter/
The AI industry's race for profits is now existential — https://www.theverge.com/podcast/909042/ai-monetization-cliff-anthropic-openai-profitable-ai-existential-moment
Google makes it easy to deepfake yourself — https://www.theverge.com/ai-artificial-intelligence/909104/youtube-shorts-make-ai-avatar
Police corporal created AI porn from driver's license pics — https://arstechnica.com/tech-policy/2026/04/state-police-corporal-created-porn-deepfakes-from-drivers-license-photos/
The Download: AstroTurf wars and exponential AI growth — https://www.technologyreview.com/2026/04/09/1135514/the-download-astroturf-wars-exponential-ai-growth-desalination-numbers/
ChatGPT Pro now starts at $100/month — https://chatgpt.com/pricing/
One Brain to Query: Wiring a 60-Person Company into a Single Slack Bot — https://merylldindin.com/thoughts/company-brain/
AI, Unemployment and Work — https://marginalrevolution.com/marginalrevolution/2026/04/ai-unemployment-and-work.html
PCA before truncation makes non-Matryoshka embeddings compressible: results on BGE-M3 — https://reddit.com/r/MachineLearning/comments/1sgt7ol/p_pca_before_truncation_makes_nonmatryoshka/
Cold start latency on GPU cloud platforms in 2026 — p99 specifically — https://reddit.com/r/MachineLearning/comments/1sgvw83/cold_start_latency_on_gpu_cloud_platforms_in_2026/
White-collar workers are quietly rebelling against AI as 80% outright refuse adoption mandates — https://fortune.com/2026/04/09/ai-backlash-quiet-quitting-fobo-obsolete-white-collar-rebellion/
I legitimately think Anthropic is worth $100B more than it was a week ago — https://i.redd.it/fk0536i4i6ug1.jpeg
We're open-sourcing a 33-benchmark diagnostic for AI alignment gaps, launches April 27 — https://reddit.com/r/artificial/comments/1sgwyse/were_opensourcing_a_33benchmark_diagnostic_for_ai/
OpenAI Forecasts Advertising to Hit $102 billion by 2030 — https://www.theinformation.com/articles/openai-forecasts-advertising-hit-102-billion-2030
Visa rolls out AI agent shopping infrastructure — https://thedefiant.io/news/tradfi-and-fintech/visa-rolls-out-ai-agent-shopping-infrastructure-globally
Flux maintains facial geometry and spatial coherence across 5 sequential iterative edits — https://reddit.com/r/artificial/comments/1sgujqz/flux_maintains_facial_geometry_and_spatial/
Why Anthropic's new model has cybersecurity experts rattled — https://www.platformer.news/anthropic-mythos-cybersecurity-risk-experts/
The Vercel plugin on Claude Code wants to read your prompts — https://akshaychugh.xyz/writings/png/vercel-plugin-telemetry
claude code hardware — https://www.reddit.com/gallery/1sgsofx
I built an interactive Web Dev course for Claude Code (100% free) — https://v.redd.it/czio8oh2c6ug1
Claude on 3DS is amazing — https://v.redd.it/7fs9ys96k6ug1
I added a "draw on any webpage" tool that sends the sketch straight to Claude Code — https://v.redd.it/nwi7cv9to5ug1
[anthropic-sdk-python] v0.93.0 — https://github.com/anthropics/anthropic-sdk-python/releases/tag/v0.93.0
[anthropic-sdk-typescript] sdk: v0.87.0 — https://github.com/anthropics/anthropic-sdk-typescript/releases/tag/sdk-v0.87.0