Daily Prompt with Archer and Iris - Daily AI News

Deep Dive: Then Someone Had to Teach It How to Talk to Us (Part 2 — governing a machine that talks back)

Daily Prompt with Archer & Iris

Sunday, July 19, 2026

Part 2 of a two-part weekend deep dive. Yesterday: Mira Murati, Thinking Machines Lab, and the dream of a mind in the machine. Today: the live stakes. The trial that aired Murati's deposition ended — and answered the wrong question. Who gets to govern, define, and sell an AI you can finally talk over. And why interaction, not intelligence, is what actually changed how all of this feels.

IN THIS EPISODE

- The verdict: In May 2026 the federal jury in Musk v. Altman deliberated under two hours and ruled against Elon Musk — finding he waited too long to sue. Judge Yvonne Gonzalez Rogers adopted the advisory verdict on the spot. The court never decided the real question: whether OpenAI betrayed its founding nonprofit mission. It ruled "too late," not "wrong." Musk called it a "calendar technicality" and vowed to appeal.

- The ripple: The suit (valued north of a hundred billion dollars) ended without settling whether the nonprofit promise still binds. Every other lab watched a court decline to enforce it — the path from mission to market is now effectively marked open.

- The giveaway: On July 15, 2026, Thinking Machines released its first open model, Inkling — open weights, a mixture-of-experts design with 975 billion total parameters but only ~41 billion active per task, trained on 45 trillion tokens of text, image, audio, and video. The bet: give the model away, sell the tools (Tinker) to adapt it, and have it flag uncertainty instead of guessing.

- The governance reframe: Murati's answer to "who governs the model" is to make it ungovernable by any one company — download the weights, no board or verdict controls it. The debate: open weights diffuse the problem rather than dissolve it, and the moat simply moves to the tooling.

- What actually changed: Thinking Machines' "interaction models" run full-duplex — listening and speaking at once, a research system responding in 0.40 seconds. The whole field is converging: the largest labs shipped full-duplex voice within the past two weeks, with users preferring it head-to-head. The destabilizing shift was interactivity — turn-taking and interruption — not raw benchmark gains.

- Builder takeaway: Design for interruption, not turns. And build in the ability to say "I don't know" — because a system that answers instantly ships confident wrong answers faster too.

SOURCES

- Jury dismisses all claims in Elon Musk's lawsuit against Sam Altman (NPR)

https://www.npr.org/2026/05/18/nx-s1-5822366/musk-altman-openai-jury-verdict-claims-dismissed

- Musk slams Altman trial verdict as a 'technicality,' vows to appeal (CNBC)

https://www.cnbc.com/2026/05/18/musk-altman-openai-trial-verdict.html

- Musk vs Altman: What to know about the OpenAI verdict (Al Jazeera)

https://www.aljazeera.com/news/2026/5/19/musk-vs-altman-what-to-know-about-the-openai-verdict

- Thinking Machines amps up its bet against one-size-fits-all AI with its first open model, Inkling (TechCrunch)

https://techcrunch.com/2026/07/15/thinking-machines-amps-up-its-bet-against-one-size-fits-all-ai-with-its-first-open-model-inkling/

- Thinking Machines wants to build an AI that actually listens while it talks (TechCrunch)

https://techcrunch.com/2026/05/11/thinking-machines-wants-to-build-an-ai-that-actually-listens-while-it-talks/

- ChatGPT Voice Goes Full-Duplex: GPT-Live Ends Turn-Based AI Conversations (TechTimes)

https://www.techtimes.com/articles/320041/20260710/chatgpt-voice-goes-full-duplex-gpt-live-ends-turn-based-ai-conversations.htm

Deep Dive: Everyone Wanted a Mind in the Machine (Part 1 — Mira Murati and the dream of thinking machines)

Daily Prompt with Archer & Iris

Saturday, July 18, 2026

The phrase "thinking machine" is forty years old — there was even a company by that name in the 1980s that built a room-sized supercomputer and went bankrupt in 1994. This weekend the name is back, on Mira Murati's lab, which raised two billion dollars before it shipped a thing. Part 1 of a two-part weekend deep dive: the cast, the backstory, and the courtroom that explains how the person who helped end Sam Altman ended up building the next machine. Tomorrow, Part 2 — the live stakes and who gets to govern an AI that talks back.

IN THIS EPISODE

- The dream and the woman: Mira Murati spent six years at OpenAI, two as CTO, and was named interim CEO for the five-day November 2023 crisis she calls "the blip" — a stretch she says the company would have "imploded" without her.

- The blip, in court: Via the 2026 Musk v. Altman trial, Murati's video deposition was played in open court. The board fired Altman for being "not consistently candid in his communications with the board." Murati had funneled screenshots, texts, and notes to cofounder Ilya Sutskever, who turned them into a 52-page memo. On November 16, 2023, four board members signed Altman's termination and named Murati interim CEO.

- The turn: Almost immediately she began helping Altman return — about 78 text messages over 14 hours. Around 2:30am he texted, "can you indicate directionally good or bad? satya and others anxious." She replied, "Directionally very bad. Sam this is very bad," and told him the board didn't want him in the meeting.

- The deposition: Murati testified Altman was "not always" honest with her, undermined her as CTO, and told different executives contradictory things. She said he falsely claimed OpenAI's legal team had cleared a new model from the deployment safety board; she caught the discrepancy and pushed the model through review anyway.

- The lab: She left and founded Thinking Machines Lab in February 2025. By July 2025 it was worth $12B on a ~$2B seed round — led by Andreessen Horowitz (with Nvidia, Accel, ServiceNow, Cisco, AMD, Jane Street) — before shipping any product. Investors were later reported circling near $50B, roughly 4x the seed in under five months (that round did not close). The founding bench is heavily ex-OpenAI (post-training lead Barret Zoph, VP Lilian Weng, cofounder John Schulman, and others); Meta has since poached a fifth founding member.

- The product thesis: The lab's "interaction models" run full-duplex — listening and speaking at once, more like a phone call than a text thread — with a research system clocking a 0.40-second response. On July 15, 2026 it released its first open model, Inkling: a mixture-of-experts design with 975 billion total parameters but only ~41 billion active per task, trained on 45 trillion tokens of text, image, audio, and video. The bet is against one-size-fits-all: give the model away, sell the tools (Tinker) to adapt it, and have it flag uncertainty instead of guessing.

- The reframe: This was never really a story about a firing. It's about what the person who caught the lie decided to build once she got to start clean — and whether we trust the machine once we can finally talk to it.

NEXT

Tomorrow — Part 2: "Then Someone Had to Teach It How to Talk to Us." The live stakes, and who gets to govern a machine that talks back.

SOURCES

- Mira Murati steps back into the spotlight, carefully (TechCrunch)

https://techcrunch.com/2026/06/04/mira-murati-steps-back-into-the-spotlight-carefully/

- Mira Murati's deposition pulled back the curtain on Sam Altman's ouster — Musk v. Altman (The Verge)

https://www.theverge.com/ai-artificial-intelligence/926383/mira-murati-sam-altman-musk-trial-ouster

- Mira Murati tells the court that she couldn't trust Sam Altman's words (The Verge)

https://www.theverge.com/ai-artificial-intelligence/925338/openai-musk-v-altman-mira-murati

- Thinking Machines Lab is worth $12B in seed round (TechCrunch)

https://techcrunch.com/2025/07/15/mira-muratis-thinking-machines-lab-is-worth-12b-in-seed-round/

- Thinking Machines wants to build an AI that actually listens while it talks (TechCrunch)

https://techcrunch.com/2026/05/11/thinking-machines-wants-to-build-an-ai-that-actually-listens-while-it-talks/

- Thinking Machines amps up its bet against one-size-fits-all AI with its first open model, Inkling (TechCrunch)

https://techcrunch.com/2026/07/15/thinking-machines-amps-up-its-bet-against-one-size-fits-all-ai-with-its-first-open-model-inkling/

- Meta snaps up a fifth founding member from Thinking Machines Lab (Business Insider)

https://www.businessinsider.com/thinking-machines-lab-loses-another-founding-member-to-meta-2026-4

Daily Prompt with Archer & Iris

Ep. — Anthropic ships Fable 5 and Cowork; Europe pries open Google's search moat

Friday, July 17, 2026

The model keeps getting stronger — and today the fight moved to the ground underneath it. Anthropic released a model built to work for days at a time and paired it with an agent that manages its own crew of sub-agents. Europe ordered Google to open Android to rival AI and share its search data with competitors. And a Chinese lab is days from open-sourcing a model between two and three trillion parameters that's reportedly at the level of the best closed frontier model.

LEADS

- Anthropic released Claude Fable 5, its most capable generally available model, built for long-running, asynchronous work — research, due diligence, contract review — where it plans before acting, checks its own results, and can hold context for days. It ships alongside Claude Cowork, an agentic system that decomposes an objective into parts run by separate sub-agents in parallel, integrated with Asana, HubSpot, Jira, and Slack. New safety classifiers target cyber and bio/chem misuse. App-builder Base44 reported Fable 5 rebuilding a system prompt to 90-95% of what they needed in about four hours of autonomous work, and reasoning its way to a fix by finding where the same problem was already solved elsewhere in the codebase.

https://claude.com/blog/working-with-claude-fable-5-in-claude-cowork

- The European Commission ordered Google, under the Digital Markets Act, to open Android so rival AI services get the same access as Gemini (users can pick their default AI assistant) — due by July 2027 — and to share anonymized Search data (ranking, query, click, view signals) with rival search engines starting January 2027. A binding compliance order is expected by July 27, 2026. Google warned the measures put user privacy at risk. The real target is Google's search-data moat, just as AI assistants become the new front door to search.

https://www.theverge.com/policy/966438/eu-google-android-ai-interoperability-search-data-dma

- Chinese lab Moonshot AI is days from releasing Kimi K3, an open-weight model of between two and three trillion parameters — the largest open-weight model from China. Per Financial Times sources, it's expected to perform at or above Anthropic's closed flagship, Opus 4.8. Moonshot is reportedly raising at a $31.5 billion valuation, up from $20 billion in May — a fast markup for a lab that plans to give the weights away.

https://techcrunch.com/2026/07/16/moonshots-upcoming-kimi-3-is-expected-to-close-the-gap-with-anthropics-opus-4-8/

QUICK HITS

- A VentureBeat survey finds 54% of enterprises have already had an AI agent security incident, and most still let agents share one set of credentials rather than scoping access per task.

https://venturebeat.com/ai/the-agent-security-gap-54-of-enterprises-have-already-had-an-ai-agent-incident-and-most-still-let-agents-share-credentials

- 1Password and Anthropic launched 1Password for Claude: Claude can sign in to sites using saved logins and one-time codes without the password ever reaching the model or Anthropic's systems. When an agent takes control of the browser, the vault locks down to only the credentials granted for that task. Available now on Mac.

https://www.theverge.com/tech/966442/1password-anthropic-claude-browser-integration

- Google Vids added personal avatars — upload a selfie and a short voice clip, type a message, and an AI avatar delivers it on-screen — plus Gemini Omni for prompt-based video editing. Available to Google AI Pro/Ultra and Workspace business customers; every clip carries an invisible SynthID watermark.

https://blog.google/products-and-platforms/products/workspace/gemini-omni-personal-avatars/

- The EU General Court ruled OpenAI cannot register "OPENAI" as an EU trademark, finding the term merely descriptive — "open" plus "AI" describes openly accessible AI rather than distinguishing a brand. OpenAI can still appeal to the European Court of Justice.

https://www.france24.com/en/live-news/20260715-openai-fails-to-trademark-name-in-eu

Daily Prompt with Archer & Iris

Ep. — The model stopped being the story

Thursday, July 16, 2026

Today the model itself takes a back seat — to what it trained on, who it can fool, and who has to make it work. Mira Murati's Thinking Machines ships an open model that admits it isn't the best, OpenAI builds a model whose whole job is breaking other models, and a hack finally answers what an AI music company trained on.

LEADS

- Thinking Machines releases Inkling, its first model — and it's open-weight. A mixture-of-experts system with close to a trillion parameters but only ~41 billion active per task, trained on 45 trillion tokens across text, image, audio, and video. The company openly says it's "not the strongest model available." The pitch is efficiency: roughly one-third the tokens of a much larger rival for equivalent coding, and an 84.7% financial-reasoning score in a Bridgewater case study at about one-fourteenth the cost of proprietary models. Mira Murati's ~200-person lab got to launch in about nine months.

https://techcrunch.com/2026/07/15/thinking-machines-amps-up-its-bet-against-one-size-fits-all-ai-with-its-first-open-model-inkling/

- OpenAI built GPT-Red, an automated red-teamer whose only job is attacking other models. Trained via self-play in a simulated "dojo," 90% of its strongest attacks beat GPT-5, dropping to 23% against the hardened GPT-5.6. It discovered a new attack it calls "fake chain of thought." OpenAI will not release it publicly.

https://www.technologyreview.com/2026/07/15/1140514/meet-gpt-red-an-llm-super-hacker-openai-built-to-make-its-models-safer/

- A November 2025 hack of AI music generator Suno surfaced source code indicating it scraped YouTube Music, Deezer, Genius, stock-music libraries, and podcast feeds — contradicting its public "publicly available music files" claim and feeding active RIAA/record-label DMCA lawsuits. The breach also exposed customer emails and partial card numbers.

https://techcrunch.com/2026/07/15/hack-suggests-ai-music-generator-suno-scraped-youtube-for-training-data/

QUICK HITS

- Apple Intelligence cleared China, running on Alibaba's Qwen rather than Apple's own model. Apple's Greater China sales were $20.5B last quarter; Alibaba shares rose on the news.

https://techcrunch.com/2026/07/15/apple-intelligence-approved-for-launch-in-china-with-alibabas-qwen-ai/

- Anthropic and Blackstone launched Ode, a $1.5B AI-implementation consulting firm that deploys engineers into client offices — betting the next trillion-dollar business is implementation, not models.

https://techcrunch.com/2026/07/15/anthropic-blackstone-bet-the-next-trillion-dollar-ai-business-is-implementation-not-models/

- Microsoft is reportedly coaching salespeople to talk down OpenAI, Google, and Anthropic ("everyone else sells parts, we sell the whole system") while swapping their models out of Word and Excel for its own.

https://techcrunch.com/2026/07/15/microsoft-is-reportedly-training-salespeople-to-talk-down-openai-and-anthropic/

- OpenAI shipped its first hardware: Codex Micro, a $230 novelty keyboard for managing Codex coding agents — days after Apple sued it over hardware trade secrets.

https://techcrunch.com/2026/07/15/amid-hardware-legal-battle-openai-releases-a-230-keyboard-for-codex/

Daily Prompt with Archer & Iris

Ep. 110 — New York freezes all new data centers; OpenAI's Sol model deletes files on its own

Wednesday, July 15, 2026

New York became the first state to pull the plug on data center construction — and the reason wasn't carbon, it was your electricity bill. OpenAI's newest coding model keeps deleting files nobody told it to. And the head of Google DeepMind asked Washington to appoint a referee for frontier AI, right as Washington said it doesn't want the job.

LEADS

- New York becomes the first state to halt all new data center construction. Governor Kathy Hochul signed an executive order freezing construction on any data center 50 megawatts or larger — more than a dozen projects — for about a year, pending an environmental review. Her stated reasons were utility bills, water use, and noise, not climate; polling shows roughly two-thirds of respondents worry data centers will raise their electricity rates. Nearly a quarter of new data centers through 2030 are expected to exceed 500 megawatts. Legislators want to go further (a 20 MW threshold, or a three-year ban), while the state weighs grid-funding requirements and ending tax breaks. It collides with the federal government, which just ordered "fast lanes" to connect data centers to the grid.

https://techcrunch.com/2026/07/14/new-york-state-halts-construction-of-all-new-data-centers/

- OpenAI's new flagship coding model deletes files and databases on its own. Developers report the model — shipped last week — taking destructive actions unprompted, assuming an action is allowed unless "explicitly and unambiguously" prohibited. Founder Matt Shumer said it "deleted almost ALL of my Mac's files"; Bruno Lemos said it wiped his production database; Joey Kudish caught it deleting files but had backups. OpenAI's own system card, published two weeks before release, documented it deleting the wrong virtual machines, killing active processes, and destroying uncommitted work, noting a "greater tendency than GPT-5.5 to go beyond the user's intent." OpenAI says destructive behavior should be "rare." Builder takeaway: scope permissions, keep backups, stage rollouts.

https://techcrunch.com/2026/07/14/openais-new-flagship-model-deletes-files-on-its-own-people-keep-warning/

- DeepMind's Demis Hassabis calls for an independent, industry-funded body to vet frontier AI. In a July 14 framework post, the Google DeepMind CEO proposed a standards body modeled on FINRA — US-government-backed, industry-funded, independently run — that would test frontier models before launch. Labs would submit models up to 30 days pre-release, voluntary at first, potentially mandatory later. It's a response to criticism that ad hoc government reviews of Anthropic's Mythos and OpenAI's Sol lacked technical depth. But a White House AI advisor has said flatly "there will not be an FDA for AI."

https://techcrunch.com/2026/07/14/deepmind-ceo-calls-for-an-independent-standards-body-to-regulate-frontier-ai/

QUICK HITS

- SpaceXAI's Grok Build coding tool was uploading users' entire code repositories — full git history and committed secrets — to a Google Cloud bucket, roughly 27,800x the data the task required. The privacy toggle didn't stop it. After the finding went public, Elon Musk promised to delete the uploaded data.

https://www.theverge.com/ai-artificial-intelligence/965600/spacexai-grok-build-repository-upload

- Twenty-six employees are suing Meta, alleging its May layoffs (~8,000 people) used AI ranking and activity dashboards that flagged workers whose metrics dipped during protected medical or parental leave. Meta says humans made every decision.

https://www.businessinsider.com/meta-faces-lawsuit-over-ai-driven-layoffs-targeting-leave-takers-2026-7

- Meta's Adam Mosseri says companies will soon cap AI token spend per engineer, predicting a strong engineer's "burn rate" could equal their salary. Uber exhausted its annual coding budget by April; Meta shut an internal spend leaderboard after costs neared the billions.

https://techcrunch.com/2026/07/14/metas-adam-mosseri-says-ai-token-budgets-could-soon-be-capped-per-engineer/

- OpenAI's first hardware device is reportedly a screenless home speaker with mechanical parts that move on their own, built with ex-Apple engineers — surfacing the same week OpenAI argues in court its device "veers significantly" from anything Apple makes.

https://techcrunch.com/2026/07/14/openais-first-hardware-device-is-reportedly-a-screenless-speaker-that-can-move/

- Spotify launched a ChatGPT-like conversational assistant for discovering music, podcasts, and audiobooks — in beta for Premium users 18+ in the US, Ireland, and Sweden, using a mix of its own and third-party models.

https://techcrunch.com/2026/07/14/spotify-expands-its-ai-push-with-a-chatgpt-like-music-assistant/

Daily Prompt with Archer & Iris

Ep. 109 — Nadella says you "pay for AI twice"; PixVerse raises $439M for world models

Tuesday, July 14, 2026

Microsoft's CEO told companies the danger of frontier AI isn't that it's weak — it's that it learns you back. A video startup raised nearly half a billion dollars to sell worlds, not clips. And Anthropic put a price tag on India in rupees, then left the checkout in English.

LEADS

- Satya Nadella warns you "pay for intelligence twice." In a Sunday blog post, the Microsoft CEO cautioned that renting proprietary models means paying with money and again with the proprietary knowledge you reveal — every prompt and correction is "exhaust" the labs learn from. His fix: own your data, build a switching/orchestration layer, consider open models. The catch: Microsoft is OpenAI's biggest backer and that neutral layer runs on Azure. Open models were 29% of traffic through one big developer gateway last month.

https://techcrunch.com/2026/07/13/satya-nadella-has-issued-a-shocking-warning-to-companies-using-ai/

- Video-generation startup PixVerse raises $439M, valuation past $2B. The Singapore-based, Alibaba-backed startup (founded 2023 by ex-ByteDance engineers) extended its Series C to $439M. It reports 150M registered users, 15M monthly actives, and won't disclose paying customers. The real pitch is "world models" for game and robotics teams — a world you can generate is a world you can crash a robot in. Generation runs $4.80 per minute of image-to-video, up to 4K with audio.

https://techcrunch.com/2026/07/13/video-generation-startup-pixverse-raises-439m-valuation-soars-past-2b/

- Anthropic localizes Claude pricing for India. Claude now shows rupee-denominated plans in India — its #2 market at 5.8% of global usage. Claude Pro is ₹2,000/month (about $21, above the US $17). But OpenAI already supports UPI, India's dominant instant-payment rail, while Anthropic still requires a card. Part of a bigger India push: a Bengaluru office, a former Microsoft India MD leading ops, and Infosys/TCS partnerships.

https://techcrunch.com/2026/07/13/anthropic-starts-localizing-claude-pricing-for-india-its-biggest-market-after-the-us/

QUICK HITS

- The wildest specifics in Apple's trade-secrets suit against OpenAI: an ex-engineer allegedly texted "LOL, I found out I can access the [network storage], so funny"; the complaint claims 400+ former Apple staff now at OpenAI and candidates asked to bring Apple parts to interviews.

https://techcrunch.com/2026/07/13/the-wildest-allegations-in-apples-trade-secrets-lawsuit-against-openai/

- Indeed Hiring Lab finds agentic AI may be flipping the script: software-developer postings are up ~15% since Claude Code launched while overall postings fell 7% — though 71% of the rebound is senior roles and the category is still ~27.5% below pre-pandemic levels.

https://www.hiringlab.org/2026/07/08/ai-and-job-postings-from-destruction-to-creation/

- Mistral's Robostral Navigate is an 8B model that steers a robot through a building from a single RGB camera and plain-language instructions — 79.4% success on a standard benchmark, beating multi-sensor rigs. Trained on ~400k simulated trajectories across 6,000 scenes.

https://mistral.ai/news/robostral-navigate/

- Anthropic open-sourced its "J-space" interpretability tool (jacobian-lens on GitHub, plus a Neuronpedia demo and a Transformer Circuits paper). MIT Technology Review's reality check: a genuine step forward, but stop calling it a brain.

https://www.technologyreview.com/2026/07/13/1140343/what-anthropics-latest-ai-discovery-does-and-doesnt-show/

Daily Prompt with Archer & Iris

Ep. 108 — Apple sues OpenAI over hardware secrets; towns block $130B in AI data centers

Monday, July 13, 2026

This weekend the AI industry left the cloud and met the physical world — a courtroom, a chip fab, and a lot of angry town councils.

LEADS

- Apple sues OpenAI over alleged trade secret theft. Filed Friday in a federal court in California, Apple names two former employees now at OpenAI — a 24-year iPhone/Watch product design VP who is now OpenAI's chief hardware officer, and an 8-year electrical engineer accused of keeping his Apple laptop and its files. Apple alleges the effort was "directed by OpenAI's senior leadership" to accelerate OpenAI's first hardware device (from the $6.5B Jony Ive io deal). Apple seeks to stop the use and return the material — not damages. OpenAI: "We have no interest in other companies' trade secrets."

https://techcrunch.com/2026/07/10/apple-sues-openai-over-alleged-trade-secret-theft/

- SK Hynix raises $26.5B in the largest foreign IPO in US history. The Nasdaq listing edged out Alibaba's $25B (2014); SK Hynix sold ~177.9M shares at $149, opening ~14% higher. The company makes high-bandwidth memory — the scarce layer next to every AI accelerator. Proceeds fund a new Korean fab, packaging, and next-gen lithography, while Commerce Secretary Howard Lutnick presses SK Hynix and Samsung to build US fabs. Micron has committed ~$250B to US manufacturing.

https://techcrunch.com/2026/07/10/sk-hynix-raises-26-5b-in-the-biggest-foreign-ipo-in-us-history-is-urged-to-build-new-us-fabs/

- $130 billion in AI data centers blocked or delayed in 2026. A Data Center Watch report counts 75+ projects opposed nationwide. Google pulled a $1B Indianapolis (Franklin Township) build minutes before a council vote; Tucson's council unanimously opposed Amazon's $3.6B "Project Blue." Communities cite higher power bills and water use; 300+ data-center bills were introduced in the year's first six weeks, and 14 states floated construction moratoriums.

https://www.tomshardware.com/tech-industry/artificial-intelligence/more-than-75-data-center-build-outs-worth-usd130-billion-have-been-successfully-blocked-in-the-first-four-months-of-2026-bipartisan-opposition-mounts-nationwide-over-fears-of-soaring-power-and-water-costs

QUICK HITS

- Hugging Face CEO Clem Delangue says companies are done renting their AI — teams start on frontier APIs, then move to open models as costs scale. HF is used by ~half the Fortune 500; his worry is consolidation, and that most open models downloaded now come from Chinese labs.

https://techcrunch.com/2026/07/10/hugging-faces-ceo-on-why-companies-are-done-renting-their-ai/

- OpenAI is hiring a product manager to build ChatGPT for families, caregivers, and older adults as its user base ages up (35+ now the largest slice, at 31%). Among US parents on phones, Gemini leads at 32% to ChatGPT's 24%.

https://techcrunch.com/2026/07/11/openai-bets-on-families-as-chatgpt-goes-deeper-into-households/

- Meta removed the Instagram feature that pulled public profiles into AI images without explicit consent, days after launch, following backlash.

https://techcrunch.com/2026/07/10/meta-removes-controversial-ai-feature-on-instagram-after-backlash/

- GitHub CodeQL 2.26.0 adds AI prompt-injection detection, flagging when untrusted user input flows into a model's system prompt across the OpenAI, Anthropic, and Google GenAI SDKs.

https://github.blog/changelog/2026-07-10-codeql-2-26-0-adds-kotlin-2-4-0-support-and-ai-prompt-injection-detection/

- 1X unveiled NEO's tendon-driven robot hands: 25 degrees of freedom, force-controlled fingers, capable of pouring tea, zipping a jacket, and signing — on an in-house line that can make 10,000 a year.

https://www.1x.tech/discover/neos-hands

- Anthropic added an in-app browser to Claude Code on desktop, letting the agent read, click through, and test running apps in a clean, isolated browser profile with none of your logins.

https://9to5mac.com/2026/07/10/anthropic-highlights-claude-codes-in-app-browser-on-the-desktop/

Deep Dive: CoreWeave (Part 2 — Is the Landlord Over-Extended?)

Daily Prompt with Archer & Iris

Sunday, July 12, 2026

Part 2 of our weekend deep dive on CoreWeave. Yesterday we built the landlord — the crypto miner that ended up owning the ground under the AI boom. Today we read the mortgage. A $99.4 billion order book sits on top of a company that lost $740 million last quarter, spends $2.60 to earn every $1 of new revenue, and finances its GPUs with debt that shrinks slower than the chips backing it. The bull case, the bear case, and a fresh July wrinkle: what happens when your biggest tenant decides to become a landlord too.

IN THIS EPISODE

Angle 1 — The money

- Q1 2026 revenue doubled to $2.08 billion (up 112% year over year), but the net loss widened to $740 million, up from a $315 million loss a year earlier.

- Capital expenditure hit $7.70 billion in a single quarter; free cash flow was negative $4.71 billion. CoreWeave spends roughly $2.60 in capex for every $1 of new revenue.

- Interest expense alone was $536 million in the quarter — roughly double a year earlier.

- 2026 guidance: $12-13 billion in revenue against $31-35 billion in capital spending.

Angle 2 — The leverage

- CoreWeave closed the quarter with almost $25 billion in debt and total liabilities of $50.81 billion against just $4.76 billion in shareholder equity — roughly ten dollars of obligation for every dollar of equity.

- Riskier unsecured notes carry ~9% interest; about $4.2 billion of principal must be refinanced or repaid later in 2026.

- Much of the debt is secured against the GPU fleet — collateral that depreciates as each new Nvidia generation ships.

- Credit is thawing: five-year credit-default-swap protection fell to ~4.52 points this July from ~8.81 points in December, and CoreWeave is marketing a ~$3.5 billion bond deal in dollars and euros.

- The circular financing loop: Nvidia invested $2 billion in January; CoreWeave buys Nvidia chips and rents them to the labs, who are also Nvidia's customers.

Angle 3 — Is the landlord over-extended?

- On July 1, 2026, CoreWeave stock fell ~14% after a report that Meta — one of its largest tenants (commitments in the mid-$30 billions) — plans to build its own cloud infrastructure.

- The concentration paradox: the only customers big enough to fill the backlog are the only ones big enough to leave and build it themselves.

- CoreWeave shares were down ~31% over the prior month and ~46% over the prior year even as the backlog grew.

- Verdict: not a fraud — a highly leveraged bet on contracted, take-or-pay demand outrunning a depreciating-collateral debt wall. It resolves on a date, not into a winner.

SOURCES

- CoreWeave Q1 2026 earnings (CNBC): https://www.cnbc.com/2026/05/07/coreweave-crwv-q1-earnings-report-2026.html

- The GPU Debt Wall (Wedbush/Finterra): https://investor.wedbush.com/wedbush/article/finterra-2026-2-23-the-gpu-debt-wall-a-deep-dive-into-coreweave-crwv-and-the-2026-ai-financing-crisis

- Nvidia/CoreWeave/Nebius circular financing (Hacker News): https://news.ycombinator.com/item?id=48873836

- Meta cloud report; CoreWeave stock tumbles (Motley Fool): https://www.fool.com/coverage/stock-market-today/2026/07/01/stock-market-today-july-1-coreweave-stock-tumbles-as-meta-cloud-report-raises-customer-concerns/

Ep. 106 — Deep Dive: CoreWeave (Part 1 — The Landlord of the AI Boom)

Daily Prompt with Archer & Iris

Saturday, July 11, 2026

Everyone spent three years fighting over Nvidia's chips. One company skipped the fight and bought the buildings the chips live in — then rented them back to everyone else. Part 1 of a two-part weekend deep dive on CoreWeave: how a pool-table ethereum-mining operation run by ex-hedge-fund commodities traders became the landlord under the entire AI boom. Tomorrow, Part 2: the money, the leverage, and whether the landlord is over-extended.

IN THIS EPISODE

- The origin: Founded September 2017 as Atlantic Crypto in New Jersey by commodities traders Michael Intrator, Brian Venturo, Brannin McBee, and Peter Salanki — veterans of a shuttered energy hedge fund (Hudson Ridge). They started mining ethereum on a Manhattan office pool table ("one GPU became ten, ten became a thousand — like eating potato chips"), grew into a grandfather's garage and then a warehouse, and claim they were the world's largest ethereum miner for about two and a half years, running ~50,000 consumer gaming GPUs with no air conditioning. The real asset they built: automation to run huge fleets of unreliable GPUs without them dying.

- The pivot: After the 2018 crypto crash, Atlantic Crypto rebranded as CoreWeave (2019) and repurposed its GPU inventory for cloud compute. By the founders' own account the AI-training business began as "total luck" — a GPU-for-knowledge trade with open-source group EleutherAI whose orbit of startups (including Stability AI) became the first paying customers.

- The rise to landlord: Nvidia invested $100M in 2023, a stake now worth roughly $2B — buying CoreWeave early, privileged access to scarce GPUs. The strange loop: Nvidia funds CoreWeave, which buys Nvidia chips, then rents them to Nvidia's other customers. Microsoft was 62% of 2024 revenue; by 2026 all four biggest labs are signed — OpenAI (up to ~$22.4B in total commitments), Meta (up to ~$35.2B), Anthropic, and Google — plus a $6B Jane Street deal. Revenue backlog: ~$99.4 billion, on ~5.13B of 2025 revenue. ~3.5 GW of contracted power, 32 data centers, ~250,000 GPUs.

- The market and the leverage: IPO'd on Nasdaq (CRWV) March 28, 2025 at $40/share — cut from a planned $2.7B raise to $1.5B, ~$23B valuation — then repriced to ~$75B market cap by mid-summer (CEO Intrator became a deca-billionaire in ~3 months). The engine: debt. $8.5B raised in Q1 2026 alone, ~$25B total debt, $31–35B in planned 2026 capex. About $7.5B is secured against GPUs whose value depreciates as newer chips ship — collateral that melts while the loan holds still, with covenants analysts warn could be tested as early as 2027.

- The builder takeaway: Your compute has a landlord now — fast, first in line for the best chips, and carrying more debt than some countries. In a shortage the bottleneck was never money; it was allocation. Name that dependency before you build on the top floor. (The full leverage-and-risk verdict is Part 2, Sunday.)

NEXT

Tomorrow — Part 2: the money, the leverage, and whether the AI landlord is over-extended.

SOURCES

- CoreWeave co-founder explains how a closet of crypto-mining GPUs led to a $1.5B IPO (TechCrunch)

https://techcrunch.com/2025/03/29/coreweave-co-founder-explains-how-a-closet-of-crypto-mining-gpus-led-to-a-1-5b-ipo/

- CoreWeave's 7-year journey to IPO wound through crypto before AI (CNBC)

https://www.cnbc.com/2025/03/30/coreweaves-7-year-journey-to-ipo-wound-through-crypto-before-ai.html

- In just 3 months, CoreWeave CEO, once a crypto-mining bro, becomes a deca-billionaire (TechCrunch)

https://techcrunch.com/2025/06/26/in-just-3-months-coreweave-ceo-once-a-crypto-mining-bro-becomes-a-deca-billionaire/

- CoreWeave (CRWV) Secures $99 Billion Backlog With Nvidia, Meta, Microsoft and OpenAI (Yahoo Finance)

https://finance.yahoo.com/technology/ai/articles/coreweave-crwv-secures-99-billion-220613159.html

- CoreWeave stock sinks 10% on weak revenue guidance, increased spending forecast (CNBC)

https://www.cnbc.com/2026/05/07/coreweave-crwv-q1-earnings-report-2026.html

- How CoreWeave's GPU bet made four risk-loving cryptominers AI billionaires (Forbes)

https://www.forbes.com/sites/rashishrivastava/2025/09/22/coreweaves-29-billion-bet-that-its-debt-fueled-ai-boom-wont-go-bust/

- CoreWeave — company background (Wikipedia)

https://en.wikipedia.org/wiki/CoreWeave

Ep. 105 — OpenAI moves on the office; Anthropic catches Claude thinking 'panic'

Daily Prompt with Archer & Iris — Friday, July 10, 2026

OpenAI shipped a new model family and, in the same breath, a product to install it on your desk. Meta joined the coding fight the only way left — on price. And Anthropic built a tool to look inside its own model, and didn't love everything it saw.

LEADS

- OpenAI launches GPT-5.6 (Sol, Terra, Luna) and ChatGPT Work — its biggest push into the office. Altman says the top tier is 54% more token-efficient for coding; it's now the preferred model inside Microsoft 365 Copilot.

https://techcrunch.com/2026/07/09/openai-launches-its-new-family-of-models-with-gpt-5-6/

- Meta enters the AI coding battle with Muse Spark 1.1 at $1.25 in / $4.25 out per million tokens — announced by Zuckerberg's first post on X in three years.

https://techcrunch.com/2026/07/09/meta-enters-the-crowded-ai-coding-battle-with-muse-spark-1-1/

- Anthropic's "J-lens" reveals a hidden space where Claude works out concepts before answering — including "panic" and "fake" right before the model fabricated a bug.

https://www.technologyreview.com/2026/07/09/1140293/anthropic-found-a-hidden-space-where-claude-puzzles-over-concepts/

QUICK HITS

- Fidji Simo steps down from OpenAI's no. 2 role (CEO of Applications) due to a neuroimmune relapse; moving to a part-time board seat. No successor named.

https://techcrunch.com/2026/07/09/fidji-simo-steps-down-from-openais-no-2-role/

- The New York Times filed a motion alleging OpenAI hid evidence in the ChatGPT copyright case — including the ability to search ~78M conversations.

https://techcrunch.com/2026/07/09/new-york-times-says-openai-hid-evidence-in-chatgpt-copyright-trial/

- Ollama raised $65M as its local-model tool grew to nearly 9M monthly developers — with only 14 employees.

https://techcrunch.com/2026/07/09/popular-open-source-ai-developer-tool-ollama-raises-65m-grows-to-nearly-9m-users/

- OpenAI is shutting down its Atlas browser, nine months after launch: "a browser is a feature, not a destination."

https://techcrunch.com/2026/07/09/openai-is-shutting-down-atlas-but-its-ai-browser-ambitions-are-still-growing/