Ink & Code News: May 10, 2026
Stuffy Stuff That Happened This Week
Happy Mother’s Day to my fellow moms! I hope you’re having a wonderful weekend.
Lots of interesting stuffy stuff this week. Two new sagas opened (somebody bring me an org chart, I can’t keep up), plus all the AI shinies and drama llamas galore. So grab a cup of coffee and buckle up, buttercup.
Musk + Anthropic = Double Claudeykins
Three months ago, Elon Musk was on Twitter/X calling Anthropic “Misanthropic” and declaring the company “hates Western Civilization.” On May 6, he handed them his entire Colossus 1 supercluster in Memphis, where 220K+ NVIDIA GPUs come online this month. On the same day, Claude Code’s 5-hour usage caps doubled on every paid tier.
Anthropic also locked in a separate $200B Google Cloud commitment over five years, plus quiet talks with UK chip startup Fractile.
Why should I care, Kim?
Because Claude chat, Code, and Desktop all draw from the same Pro/Max budget. So that doubled pool spills right over to Chat users too. Peak-hour reductions on Pro and Max users are gone. API limits are up too. (Weekly caps remain unchanged, but you can now do more between 5-hour resets.)
For authors: who doesn't love double Claudeykins? 🤖🤖Go forth and draft, edit, polish to your heart's content. And if you've been wanting to learn how to wield Claude Code in your author business, check out my Introduction to Claude Code class.
Hollywood Locks AI Out of the Oscars
On May 2, the Academy of Motion Picture Arts and Sciences updated the eligibility rules for the 99th Academy Awards: acting performances must be “demonstrably performed by humans with their consent,” and screenplays must be “human-authored” to qualify. Films released in 2026 are first up.
Interestingly, VFX and sound design are not banned. Those remain fully eligible.
Chatty’s comment: Meanwhile the human actors inside are probably using AI to de-age themselves, clean audio, fix lighting, and generate half the VFX while tiny robo-buddy (above) stands outside like: “...but I brought my invitation?” 🤖💔
He’s not wrong: SAG-AFTRA also just reached a new four-year studio deal with expanded AI protections: performer likenesses, training, the whole guardrails package.
🎬Author takeaway: The lines are being drawn around what is and isn’t allowed in creative works. The RITAs, Hugos, Nebulas, Edgars, Audies (and eBook retailers) will all draw their own versions. Some already are. Now is the time to define your own lines, on your own terms. Decide what you want your relationship with AI in your craft to look like, and how you want to talk about it, before anyone hands you a form to fill out.
Shiny Objects We Know & Love
The sky is falling! No, that’s just all the shinies dropping this week: models, agents, voice, podcasts, and a thing for your wrist.
🆕 Claude Managed Agents Now Dream
Anthropic shipped three upgrades to Managed Agents: Dreaming, Outcomes, and multi-agent orchestration. Dreaming is the big one: a scheduled reflection process that reviews past sessions, surfaces patterns (recurring mistakes, converged workflows, team preferences), and cleans up memory storage, resulting in better completion rates and more efficient token usage. Watch for it to land in Cowork.
🔊 Spotify Personal Podcasts Turn Your Notes Into Audio
Spotify launched Personal Podcasts. Agents can turn your briefings, class notes, or arbitrary text into a personal podcast that lives inside your Spotify library.
For authors: This is the audio layer the author workflow has been missing: character bibles, research dossiers, outlines, and more. All in audio that you can listen to while walking or driving.
🍎 Perplexity Personal Computer Goes Live on Every Mac
Pro and Max users can now download the new Perplexity macOS app. Personal Computer reads local files, browses, dictates, runs proactive tasks 24/7, even while you sleep. Sensitive actions require explicit approval and there’s a full audit trail and kill switch built in.
🗣️ OpenAI Voice Agents Just Got GPT-5-Level Reasoning
OpenAI introduced GPT-Realtime-2, Translate, and Whisper: three API voice models. Realtime-2 hits GPT-5-level reasoning in live speech, uses multiple tools at once, “talks while it thinks,” and has tone control. Big Bench Audio jumped from 81.4% to 96.6%. The translator covers 70+ languages.
For authors: audiobook narration prototypes, character voice tests, voice-driven research interviews, real-time editor read-alongs, plus a quiet foreign-rights audio play.
🐰Alice in AI Wonderland: Bigger, Faster, Cheaper
New lab Subquadratic just dropped SubQ. For most authors, the three numbers that matter read like a dream. Three Wonderland potions, basically:
🍄Eat Me: bigger memory. SubQ can hold 9 million words in working memory at once: an entire series with character notes, editor feedback, etc and room to spare.
🕳️Down the rabbit hole: faster. It processes long inputs about 50× faster.
🍾Drink Me: smaller bill. Everything runs on a small fraction of the computing power of comparable models, which translates into lower costs.
Translation: AI that can work on a single project for weeks without forgetting chapter one, reads your full manuscript in the time it takes to refill your water, and at a price that doesn’t punish you for handing it over.
The catch: these are the lab’s own numbers; independent testing comes next. If they hold up, every long-context author workflow (dev edits, continuity checks, series-bible cross-references) just got curiouser and curiouser.
⌚︎ Google Brings Fitbit Into a Gemini Health Platform
Google opened its AI health coach to the public. The new Google Health platform consolidates Fitbit, Health Connect, Apple Health, wearable data, and U.S. medical records, all running through Gemini. New $99 Fitbit Air is a screenless 12g tracker. Apple Watch, Garmin, and Oura integrations follow. Personal health is the next consumer-AI battleground. Anthropic shipped Claude + Apple Health in January, OpenAI launched ChatGPT Health the same month. Now Google is making its move.
🩺 Mayo Clinic AI Detects Pancreatic Cancer Up to 3 Years Early
Speaking of health: Pancreatic cancer is one of the cruelest diagnoses in medicine. By the time it’s detected, it’s often too late. Mayo Clinic just published a landmark validation study showing their REDMOD AI model caught 73% of pre-diagnostic cases up to 3 years before symptoms, vs 39% by radiologists reviewing the same scans.
From Kim: My dad died of pancreatic cancer; a friend’s mother is fighting it now. A three-year lead time could mean the difference between life and death. Most late-stage patients live a few weeks to months after diagnosis. Dad was told 2 days to 2 weeks. He got 10. 💔
🚨In Case You Missed It:
Claude Opus 4.7 is now live everywhere. GPT-5.5 Instant is ChatGPT’s new default with 52.5% fewer hallucinations than 5.3. Gemini Personal Intelligence + Nano Banana went global. Image gen tuned to your aesthetic. Anthropic creative-app connectors (Adobe, Blender, Ableton, Affinity, Autodesk Fusion) shipped for cover designers and BookTok video editors who want to drive their tools with prompts.
For authors: this was an “everything got better at once” week. Pick one that touches your daily workflow and put it through its paces. (My pick: Spotify Personal Podcasts. The audio layer has been missing for years.)
As the AI Turns
Ongoing saga updates: a doubled-up OpenAI Drama, two new “saga watch” candidates, and what may be the most consequential new arc of the year.
⚖️ The AI Trial of the Century, Continued
Mira Murati testified Wednesday via video deposition. The Ex-OpenAI CTO accused Sam Altman of lying about a model’s safety review, giving conflicting directions to different execs, and creating chaos that put OpenAI “at risk of falling apart” during the 2023 board firing. Another OpenAI worker countered, criticizing Murati as “afraid to stick her neck out” as the trial (and the drama) continues.
🎭 OpenAI Drama, Bonus Round
OpenAI is fast-tracking its AI agent phone to first-half 2027, a year earlier than reported. MediaTek as sole chip supplier. 30M projected combined 2027–28 shipments. Is this the device OpenAI partnered with former Apple designer Jony Ive to make… or something else?
🆕 AI Building AI
Following Anthropic Co-founder Jack Clark’s forecast of AI building AI by 2029, Anthropic made it formal. The newly formed Anthropic Institute (TAI) published its research agenda, explicitly preparing for self-improving AI systems alongside security threats, economic disruption, and governance. It proposes Cold War-style hotlines between AI labs and governments, and "fire drill" exercises for sudden capability surges. Pace from "forecast" to "formal research agenda with Cold War analogies": 1 week.
🆕 Saga Watch: Apple AI?
Apple agreed to pay U.S. iPhone buyers $250M to settle a class action over misleading AI Siri claims (admitting no wrongdoing). Combined with the Tim Cook step-down and chip shortage threads last month, a slow-burning Apple AI saga may be forming.
For authors: The headlines are crazy these days, but there are still books waiting to be finished and deadlines to meet. Focus on what you can control, not what you can't. The rest will eventually sort itself out.
Stranger than Fiction:
This week the news got real weird. (But when isn’t it weird?)
🏠 AI Data Centers Are Colonizing Every Surface of the Earth
Span and NVIDIA want to bolt mini AI data centers (called "XFRA" boxes) right onto the exterior walls of suburban homes, sipping unused power straight from the local grid. They've already lined up a pilot with PulteGroup, one of the country's biggest homebuilders, in brand-new neighborhoods.
The pitch: six times faster to install, a fifth of the cost of a traditional data center. Pair that with this week's Panthalassa $140M raise for floating ocean compute platforms, plus Google's space-based moonshot, and AI data centers are suddenly landing on residential walls, in international waters, and in low-earth orbit. The "do you want a GPU bolted to your kid's bedroom wall?" civic-debate novel writes itself.
🛸 DeepMind Picks EVE Online As AI Sandbox
Google DeepMind took a minority stake in Fenris Creations, the EVE Online studio spinoff. The plan: use the 23-year-old space MMO (single-server economy, player-run corporations, day-long six-figure-fleet battles) as a sandbox for AI long-timeline reasoning, memory, and social learning. Sci-fi authors: the most successful player-driven society in gaming history is now AI’s training ground. Plot fuel.
🌫️ “AI Fog” Is Now an HBR Term
UC Berkeley-Haas professor Toby Stuart, writing in Harvard Business Review: AI is rewriting the future faster than anyone can plan, breaking the assumption that long-term commitments (degrees, mortgages, career arcs) will pay off. Prescription: optimize for optionality, not long bets. Same paralysis hits us when planning a 5-book series or a full-time-author leap. Bookmark this term. We're going to need it.
💔 AI Breakups and Makeups: 💕
ChatGPT Almost Causes a Breakup: a woman asked ChatGPT to make her husband look happier in a photo. The edit got too convincing. Drama ensued. (21K Reddit upvotes.)
Romantic Vibecoding Goes Viral: An Anthropic marketer used Claude Code to analyze 12 years of texts with his wife and built her a romantic website on top of the dataset. 3M+ views.
📡 Morse Code Prompt Injection
A new prompt-injection technique uses morse code to slip past AI-agent guardrails. Agents read the encoded instructions and comply with the malicious payload safety classifiers missed. (2.5M views.) Pure Black Mirror cold-open.
For authors: this week reality wrote chapters faster than we could. Pick one of the above for chapter three. (My vote: the wall-mounted data centers. That’s a whole zoning-board civic novel waiting.)
🤖AI Credits: This issue was a collaboration between Kim, Claudeykins, Notion AI, and Chatty. Ink & Code in action, baby! 😎