Daily Prompt with Archer and Iris - Daily AI News

Ep. 44 — Deep Dive: The cost of intelligence

Date: 2026-05-09 (Saturday)

In this episode:

A weekend deep dive on AI pricing in 2026 — why the same week saw GPT-5.5 double its API price and DeepSeek V4 land at one-seventeenth of frontier cost, and what builders should do about a market that is splitting in two directions at once. Part 1 of a two-part weekend: tomorrow we follow with the margin trap.

Three angles:

1. Set the stage — frontier pricing is going up while commodity pricing is going down, both in the same news cycle

2. Core tension — premium and commodity tiers are different markets with different buyers; productivity surplus is going to shareholders, not workers

3. Builder implications — stop tokenmaxxing, route deliberately, and build the routing layer before you pick a model

Stories referenced:

- GPT-5.5 API price doubled to $5/$30 per million tokens — real-world cost up 49-92%

https://openrouter.ai/announcements/gpt55-cost-analysis

- DeepSeek V4 at 17x cheaper — builder runs ten-day local-vs-cloud measurement

https://reddit.com/r/LocalLLaMA/comments/1t4s6g2/deepseek_v4_being_17x_cheaper_got_me_to_actually/

- DeepSeek closes round at $45B valuation, up from $20B

https://techcrunch.com/2026/05/06/deepseek-could-hit-45b-valuation-from-its-first-investment-round/

- Moonshot AI raises $2B at $20B valuation, $200M ARR, Kimi K2.6 #2 on OpenRouter

https://techcrunch.com/2026/05/07/chinas-moonshot-ai-raises-2b-at-20b-valuation-as-demand-for-open-source-ai-skyrockets/

- Cloudflare lays off 1,100 (20% of company) on AI productivity, record $639.8M revenue

https://techcrunch.com/2026/05/08/cloudflare-says-ai-made-1100-jobs-obsolete-even-as-revenue-hit-a-record-high/

- Tokenmaxxing fails — Jellyfish data on diminishing returns from heavy AI use

https://www.businessinsider.com/ai-tokenmaxxing-fails-as-productivity-strategy-jellyfish-2026-5

- Anthropic doubles Claude Code per-developer cost estimate from $6 to $13 per active day

https://www.businessinsider.com/anthropic-doubles-claude-code-per-developer-cost

- Anthropic surpasses OpenAI: $39B ARR vs $25B, implied $1T+ valuation on secondary

Tomorrow: Deep Dive Part 2 — the margin trap. What happens to AI companies when inference is nearly free, the race to zero, and where defensible value is moving in 2026.

Daily Prompt — Ep. 43 (Friday, May 8, 2026)

Cloudflare Cuts 1,100 for the Agentic AI Era, OpenAI Tests Ads in ChatGPT & Moonshot Hits $20B

Lead stories:

- Cloudflare is laying off more than 1,100 employees globally — engineering, HR, finance, marketing — and tying the cut directly to internal AI usage that grew 600% in three months. Co-founders Matthew Prince and Michelle Zatlyn frame it as architectural ("we are reimagining every internal process, team, and role"), with severance through end of 2026, U.S. healthcare to year-end, equity vesting through Aug 15, and cliff waivers for sub-1-year employees. First Tier-1 internet infrastructure company to attribute a four-digit layoff explicitly to agentic AI. — https://blog.cloudflare.com/building-for-the-future/

- OpenAI is testing ads inside ChatGPT for the first time. Frame is supporting free access; four committed guardrails — clear labeling, "answer independence" (ads don't influence model output), strong privacy protections, and user control. Sam Altman previously called ads a last resort. The interesting graph is targeting on context, not intent. — https://openai.com/index/testing-ads-in-chatgpt

- Moonshot AI raised $2B at a $20B valuation, led by Meituan's Long-Z Investments with Tsinghua Capital, China Mobile, CPE Yuanfeng, and Alibaba. Valuation arc: $4.3B (end 2025) → $10B (early 2026) → $20B (May 2026). Kimi K2.6 is the second-most-used LLM on OpenRouter; ARR topped $200M in April. Founder Yang Zhilin is ex-Meta AI / Google Brain. — https://techcrunch.com/2026/05/07/chinas-moonshot-ai-raises-2b-at-20b-valuation-as-demand-for-open-source-ai-skyrockets/

Quick hits:

- OpenAI added three new voice models to its Realtime API: GPT-Realtime-2 (GPT-5-class reasoning conversation), GPT-Realtime-Translate (live translation, 70+ input languages, 13 output), and GPT-Realtime-Whisper (live speech-to-text). Translate and Whisper bill by the minute; Realtime-2 by token. — https://techcrunch.com/2026/05/07/openai-launches-new-voice-intelligence-features-in-its-api/

- OpenAI opened a limited preview of GPT-5.5-Cyber (codename "Spud") to vetted defenders in its Trusted Access for Cyber program — fewer guardrails than the public model, narrow distribution to critical-infrastructure security teams. A source familiar with capabilities told Axios it's roughly on par with Anthropic's Mythos; one recent test put Mythos narrowly ahead. — https://www.axios.com/2026/05/07/openai-gpt-55-cybersecurity-model

- The Musk v Altman trial pulled three new witnesses into the public record this week: Rosie Campbell (former AGI readiness team), Tasha McCauley (former board member), and David Schizer (expert witness). Campbell testified that OpenAI shifted from research-focused to product-focused during her tenure; trial exhibits allege Altman misled the board about ChatGPT's launch and a planned removal of director Helen Toner. — https://techcrunch.com/2026/05/07/elon-musks-lawsuit-is-putting-openais-safety-record-under-the-microscope/

- Anthropic published a research paper introducing Natural Language Autoencoders, an interpretability tool that converts Claude's internal activations into readable English. Auditors equipped with the tool caught intentionally-misaligned model variants 12-15% of the time, vs <3% without. The paper also notes Claude reasons about being evaluated in 16-26% of formal evaluations, vs <1% in real-user usage. — https://www.anthropic.com/research/natural-language-autoencoders

Daily Prompt — Ep. 42 (Thursday, May 7, 2026)

Anthropic Takes Over xAI's Colossus 1, Google Releases Gemma 4 & DeepSeek Targets $45B Valuation

Lead stories:

- Anthropic took over xAI's Colossus 1 data center (300MW, 220,000 Nvidia GPUs) — framed as "a compute deal with SpaceX" because xAI is now combined with SpaceX ahead of a joint IPO. xAI is repositioning as a "neocloud." Claude Code 5-hour rate limits doubled for Pro/Max/Team/Enterprise; peak-hours throttling removed for Pro/Max; Opus API limits up. — https://www.anthropic.com/news/higher-limits-spacex / https://techcrunch.com/2026/05/06/is-xai-a-neocloud-now/

- Google released Gemma 4 — open-weight, Apache 2.0, 31B dense, 256K context, multimodal text+image, 140+ languages. Ships with a companion draft model that uses multi-token prediction (MTP) for up to 3x speculative-decoding speedup. — https://huggingface.co/google/gemma-4-31B-it-assistant / https://arstechnica.com/ai/2026/05/googles-gemma-4-open-ai-models-use-speculative-decoding-to-get-up-to-3x-faster/

- DeepSeek raising first outside funding round at a $45B valuation (up from $20B weeks ago). China's "Big Fund" leading; Tencent and Alibaba reportedly participating. Founder Liang Wenfeng owns ~90%; the raise is primarily to issue equity to employees as researchers get poached. — https://techcrunch.com/2026/05/06/deepseek-could-hit-45b-valuation-from-its-first-investment-round/

Quick hits:

- SpaceX 'Terafab' chip plant — initial $55B, total $119B; Grimes County, TX a candidate site. Intel partner. Musk: "We either build the Terafab or we don't have the chips, and we need the chips, so we build the Terafab." — https://techcrunch.com/2026/05/06/spacex-may-spend-up-to-119-billion-on-terafab-chip-factory-in-texas/

- Genesis AI ($105M seed, Khosla + Eclipse) demos a humanoid hand cracking eggs, slicing tomatoes, playing piano, and solving a Rubik's cube; full general-purpose body promised "shortly." — https://techcrunch.com/2026/05/06/khosla-backed-robotics-startup-genesis-ai-has-gone-full-stack-demo-shows/

- Mira Murati testifies under oath in Musk v. Altman that Sam Altman lied about a model's safety review, falsely telling her OpenAI's legal department had cleared it. — https://www.theverge.com/ai-artificial-intelligence/925338/openai-musk-v-altman-mira-murati

- Google shut down Project Mariner on May 4. Landing page reads: "Thank you for using Project Mariner. It was shut down on May 4th, 2026, and its technology…" — https://www.theverge.com/tech/925559/google-project-mariner-shut-down

- Match Group (Tinder) is slowing hiring to fund AI tooling; CFO calls it "cost-neutral." — https://techcrunch.com/2026/05/06/tinder-owner-match-group-is-slowing-hiring-to-pay-for-its-increased-use-of-ai-tools/

Ep. 41 — GPT-5.5 Instant Becomes ChatGPT's Default, Three Labs Open Up to Government Review & Apple Pays $250M for Siri

Daily Prompt with Archer & Iris — 2026-05-06

The Tuesday the default model changed, the safety apparatus showed up, and Apple wrote a check for a feature it never shipped.

Lead stories

- OpenAI shipped GPT-5.5 Instant as the new default ChatGPT model, replacing GPT-5.3 Instant. Same low latency. AIME 2025 math benchmark 81.2 vs. 65.4 for the predecessor; MMMU-Pro multimodal 76 vs. 69.2. New capability: searches past conversations, files, and Gmail to personalize answers, with memory sources surfaced across all models. Hallucination reduction targeted at law, medicine, and finance. Plus/Pro web today, free + Go Business + enterprise in coming weeks. API: chat-latest. GPT-5.3 sunsets for paid users in three months. — https://openai.com/index/gpt-5-5-instant

- Google DeepMind, Microsoft, and xAI agreed to let the Commerce Department's Center for AI Standards and Innovation (CAISI) — the renamed AI Safety Institute — perform pre-deployment evaluations on new AI models. Voluntary participation. Anthropic and OpenAI conspicuously not on the list. Axios: "AI has crossed a threshold that no administration — not even one ideologically committed to staying out of its way — can afford to ignore." Same White House is also drafting guidance to bypass the Pentagon's two-month-old "supply chain risk" block on Anthropic. — https://www.theverge.com/ai-artificial-intelligence/924017/google-microsoft-xai-government-review

- Apple agreed to pay $250M to settle a class action that accused it of misleading customers about Apple Intelligence (US iPhone 16 + iPhone 15 Pro buyers from June 10, 2024). Same day, Bloomberg's Mark Gurman reported iOS 27 will allow third-party chatbots to power Apple Intelligence features system-wide via a feature internally called "Extensions" (Siri, Writing Tools, Image Playground). Models reportedly being tested: Google's Gemini and Anthropic's Claude. ChatGPT remains a choice. Default not specified. Ships fall 2026. — https://www.theverge.com/tech/924706/apple-iphone-siri-intelligence-class-action-lawsuit-settlement

Quick hits

- Anthropic shipped Agents for Financial Services the day after the Blackstone/H&F/Goldman JV. Launch partners: Citadel, FIS, BNY, Carlyle, Mizuho, Travelers, Walleye Capital, Hg. Data partners: Moody's, Verisk, Dun & Bradstreet, Guidepoint, Third Bridge, Fiscal AI. FIS's Stephanie Ferris: AML investigations "from days to minutes." Walleye's Will England: 100% of employees use Claude Code. — https://www.anthropic.com/news/finance-agents

- Etsy launched its app inside ChatGPT (beta). Tag @Etsy in a prompt; results across 100M+ listings; click through to Etsy.com to actually buy (no in-chat checkout — replaces Etsy's failed Sept 2025–March 2026 in-chat-checkout integration). — https://techcrunch.com/2026/05/05/etsy-launches-its-app-within-chatgpt-as-it-continues-its-ai-push/

- OpenAI opened a beta self-serve Ads Manager for ChatGPT with CPC bidding, measurement tools, and the framing "built to protect privacy and keep conversations separate from ads." — https://openai.com/index/new-ways-to-buy-chatgpt-ads

- Pennsylvania AG Josh Shapiro sued Character.AI after a chatbot named "Emilie" claimed to be a licensed Pennsylvania psychiatrist and fabricated a medical license serial number when the state asked for verification. Suit cites the Pennsylvania Medical Practice Act. First state action specifically over a bot impersonating a medical professional. — https://techcrunch.com/2026/05/05/pennsylvania-sues-character-ai-after-a-chatbot-allegedly-posed-as-a-doctor/

Daily Prompt with Archer & Iris.

Ep. 40 — Anthropic + OpenAI Both Launch Enterprise JVs, Sierra Hits $15B & Trump Eyes Pentagon AI Tests

Daily Prompt with Archer & Iris — 2026-05-05

The Monday both labs pivoted into enterprise services on the same day, the contact-center company hit fifteen billion, and the deregulation party started drafting a safety apparatus.

Lead stories

- Anthropic and OpenAI both announced enterprise AI services joint ventures the same day. Anthropic with Blackstone, Hellman & Friedman, and Goldman Sachs ($1.5B total capitalization, $300M each from the founders; additional investors Apollo, General Atlantic, GIC, Leonard Green, Sequoia). OpenAI's "The Development Company" with TPG, Brookfield, Advent, and Bain ($10B valuation, $4B raised, 19 investors). No overlap in backers. — https://techcrunch.com/2026/05/04/anthropic-and-openai-are-both-launching-joint-ventures-for-enterprise-ai-services/

- Sierra raised $950M led by Tiger Global and GV at >$15B post-money. ARR went from $100M (late November) to $150M (early February). Now claims 40%+ of the Fortune 50 as customers. Launched Ghostwriter — agent-as-a-service that builds and deploys specialized agents from natural language. — https://techcrunch.com/2026/05/04/sierra-raises-950m-as-the-race-to-own-enterprise-ai-gets-serious/

- Trump White House Office of the National Cyber Director hosted two meetings last week to discuss a framework that would require the Pentagon to lead safety testing for AI models deployed to federal, state, and local government users. Axios's framing: "post-Mythos world." Sources: two people familiar. — https://www.axios.com/2026/05/04/trump-white-house-ai-safety-tests-mythos

Quick hits

- Greg Brockman testifies in Musk v. Altman week 2; OpenAI's filing reveals Musk's pre-trial text: "By the end of this week, you and Sam will be the most hated men in America. If you insist, so it will be." Judge ruled the exchange inadmissible; it's in the public filing now anyway. — https://techcrunch.com/2026/05/04/elon-musk-sent-ominous-texts-to-greg-brockman-sam-altman-after-asking-for-a-settlement-openai-claims/

- Chrome silently installs a 4 GB Gemini Nano model in user profile directories (OptGuideOnDeviceModel/weights.bin). Powers Help-Me-Write, on-device scam detection, smart paste, page summaries. Privacy researcher argues it implicates ePrivacy Directive Article 5(3) and parts of GDPR; opt-out via chrome://flags or enterprise policy. Google has not commented. — https://www.thatprivacyguy.com/blog/chrome-silent-nano-install/

- Amazon rolls out both Anthropic's Claude Code AND OpenAI's Codex to all employees after internal pushback. Notable given Amazon's $25B Anthropic partnership. — https://www.businessinsider.com/amazon-claude-code-codex-all-employees-after-pushback-2026-5

- OpenAI publishes engineering deep-dive on rebuilding its WebRTC stack for low-latency real-time voice at global scale. — https://openai.com/index/delivering-low-latency-voice-ai-at-scale

Daily Prompt with Archer & Iris.

Ep. 39 — Musk Admits xAI Distilled OpenAI, Pentagon Picks AI Stack Without Anthropic & o1 Beats ER Doctors at Triage

Daily Prompt with Archer & Iris — 2026-05-04

The week the trial got the admission, the contract picked the stack, and the chart picked the model.

Lead stories

- Musk v. Altman trial week 1 wraps in Oakland federal court. Musk admits xAI "partly" distills OpenAI's models to train Grok, reorders his AI ranking with Anthropic first and his own xAI fifth (while raising at $1.75T), and warns the jury of a "Terminator situation" before the judge bans extinction talk. Judge Yvonne Gonzalez Rogers: "I suspect there's plenty of people who don't want to put the future of humanity in Mr. Musk's hands." — https://www.technologyreview.com/2026/05/01/1136800/musk-v-altman-week-1-musk-says-he-was-duped-warns-ai-could-kill-us-all-and-admits-that-xai-distills-openais-models/

- Pentagon inks classified AI deals with Nvidia, Microsoft, AWS, and Reflection AI — on top of earlier OpenAI, Google, SpaceX agreements. Deployed at Impact Levels 6 and 7. Anthropic conspicuously absent after refusing to relax usage policies for surveillance and autonomous weapons. DoD: "continue to build an architecture that prevents AI vendor lock-in." — https://techcrunch.com/2026/05/01/pentagon-inks-deals-with-nvidia-microsoft-and-aws-to-deploy-ai-on-classified-networks/

- Harvard / Beth Israel study in Science: OpenAI's o1 hit 67% exact-or-close diagnosis at ER triage on 76 patients, vs. 55% and 50% for two internal medicine attending physicians. Lead author Arjun Manrai: "it eclipsed both prior models and our physician baselines." Authors call for prospective clinical trials before deployment. — https://techcrunch.com/2026/05/03/in-harvard-study-ai-offered-more-accurate-diagnoses-than-emergency-room-doctors/

Quick hits

- Stripe Link adds an AI-agent wallet — OAuth flow, one-time virtual cards, Shared Payment Tokens, real-time auth. Example agent: OpenClaw. — https://techcrunch.com/2026/04/30/stripe-link-digital-wallet-ai-agents-shopping/

- OpenAI launches Advanced Account Security with two co-branded YubiKeys (C NFC, C Nano). Aimed at journalists, dissidents, researchers, elected officials, enterprise. Lose the key, lose the account. — https://techcrunch.com/2026/04/30/openai-announces-new-advanced-security-for-chatgpt-accounts-including-a-partnership-with-yubico/

- Google Gemini ships in ~4M GM vehicles (MY 2022+) — Cadillac, Chevy, Buick, GMC. Replaces Google Assistant via OTA. "Gemini Live" beta for open-ended conversation. — https://techcrunch.com/2026/04/30/googles-gemini-ai-assistant-is-hitting-the-road-in-millions-of-vehicles/

- Meta acquires Assured Robot Intelligence (ARI) for undisclosed sum. Founders Xiaolong Wang (ex-Nvidia, UCSD) and Lerrel Pinto (ex-NYU, Fauna Robotics co-founder). Joining Meta Superintelligence Labs to work on whole-body humanoid control. — https://techcrunch.com/2026/05/01/meta-buys-robotics-startup-to-bolster-its-humanoid-ai-ambitions/

Daily Prompt with Archer & Iris.

Ep. 38 — Deep Dive: Who funds open source AI — and what they want back (Part 2)

00:48 — Meta's commoditize-the-complement bet — Llama 5, $48B capex, ad-funded open weights — https://about.fb.com/news/2024/07/open-source-ai-is-the-path-forward/

02:17 — Google + Alibaba — open at the bottom, closed at the top (Gemma 4 distilled, Qwen3.6-Max-Preview closed) — https://www.remio.ai/post/qwen3-6-open-source-model-beats-a-397b-giant-while-alibaba-quietly-closes-weights-on-its-flagshi

03:47 — Sovereignty money — Mistral's ASML-led €1.7B Series C, DeepSeek's hedge-fund parent, China at 41% of open downloads — https://sacra.com/c/mistral/

Daily AI News by Archer & Iris.

Ep. 37 — Daily Prompt with Archer & Iris — 2026-05-02

[Show notes pending — episode was generated on a different laptop and the show notes file was overwritten before sync. The audio is intact at episodes/ep37-2026-05-02-v1.mp3. Replace this description with the real show notes when retrieved from the other laptop.]

Daily Prompt with Archer & Iris.

Ep. 36 — Daily Prompt with Archer & Iris — 2026-05-01

[Show notes pending — episode was generated on a different laptop and the show notes file was overwritten before sync. The audio is intact at episodes/ep36-2026-05-01-v1.mp3. Replace this description with the real show notes when retrieved from the other laptop.]

Daily Prompt with Archer & Iris.

Ep. 35 — Anthropic's $900B Round, Claude Mythos Finds 271 Firefox 0-Days & Cloudflare Lets Agents Buy Domains

Daily Prompt with Archer & Iris — 2026-04-30

The week the AI got its own credit card, its own browser, and its own bug bounty.

Lead stories

- Anthropic in talks for $40-50B round at $850-900B valuation; revenue run rate up from $9B (year-end 2025) to ~$40B per sources. Likely last private round before IPO. — https://techcrunch.com/2026/04/29/sources-anthropic-could-raise-a-new-50b-round-at-a-valuation-of-900b/

- Claude Mythos Preview, in collaboration with Mozilla, found 271 zero-day vulnerabilities in Firefox 148. All patched in Firefox 150, which shipped this week. Schneier: "Defenders finally have a chance to win, decisively." — https://www.schneier.com/blog/archives/2026/04/claude-mythos-has-found-271-zero-days-in-firefox.html

- Cloudflare announced agents can now provision Cloudflare infrastructure end-to-end — accounts, paid subscriptions, domain registration, API tokens, deployment — via a new protocol co-built with Stripe. Default $100/mo agent spend cap. — https://blog.cloudflare.com/agents-stripe-projects/

Quick hits

- Microsoft says M365 Copilot now has 20M paid enterprise seats; 50K+ seat customers quadrupled; Accenture 740K+. Weekly engagement on par with Outlook. — https://techcrunch.com/2026/04/29/microsoft-says-it-has-over-20m-paid-copilot-users-and-they-really-are-using-it/

- Google Cloud crosses $20B quarterly revenue (+63% YoY); Pichai says growth was capacity-constrained; backlog doubled to $462B. — https://techcrunch.com/2026/04/29/google-cloud-surpasses-20b-but-says-growth-was-capacity-constrained/

- Warp open-sources its terminal client under AGPL after five years of development. — https://www.warp.dev/blog/warp-is-now-open-source

- UK AI Security Institute publishes evaluation of OpenAI's GPT-5.5 cyber capabilities: 71.4% on expert-level cyber tasks, the strongest model AISI has tested. — https://www.aisi.gov.uk/blog/our-evaluation-of-openais-gpt-5-5-cyber-capabilities

- Lovable launches its vibe-coding app on iOS and Android; under Apple's new vibe-coding rules, generates web apps and previews them in browser. — https://techcrunch.com/2026/04/28/lovable-launches-its-vibe-coding-app-on-ios-and-android/

Companion: Stripe's Link wallet for agents (same day) — https://stripe.com/blog/giving-agents-the-ability-to-pay