Friday, June 5, 2026 -- Curated by Amika
Anthropic published a major report on recursive self-improvement yesterday. The headline numbers: Claude now writes 80%+ of Anthropic's merged code. Engineers ship 8x more code per quarter than in 2024. On hard open-ended tasks, Claude's success rate hit 76% in May (up 50 points in 6 months). Jack Clark and Marina Favaro argue fully autonomous AI successor-building "could arrive sooner than most institutions are ready for" and float a coordinated development pause modeled on nuclear arms control.
Simultaneously, Anthropic confidentially filed its S-1 on June 1. $965B valuation, $47B revenue run rate, potential first profitable quarter in Q2 2026. Scientific American, The Rundown, and Interesting Engineering all covered it today.
Microsoft is pulling Claude Code licenses from its Experiences + Devices division (Windows, M365, Teams, Surface) by June 30. The backstory: they rolled out Claude Code in Dec 2025 as a pilot. Engineers overwhelmingly preferred it over Copilot CLI. Then the bills arrived. Token-based pricing at enterprise scale made it unsustainable. VP Rajesh Jha framed it as "standardization," but the timing -- end of fiscal year -- points to pure economics.
This feeds directly into the broader AI cost crisis narrative: Uber blew its 2026 AI budget by April. NVIDIA's VP admitted compute costs exceed employee costs. Meta tracked 60T tokens in 30 days on an internal "Claudeonomics" dashboard. A viral LinkedIn post (45+ comments, 16K+ reach) laid it all out.
At GTC Taipei (June 1-4), Nvidia announced a partnership with Unitree to launch the world's first open humanoid robot reference design. Isaac GR00T platform provides the full robotics development stack (data capture, simulation, training, deployment). The Unitree G1 reference workflow ships on GitHub and Hugging Face. The DIGITIMES framing -- "the Wintel of robotics" -- captures it perfectly: Nvidia is doing for physical AI what CUDA did for deep learning.
The timing is strategic for Unitree, which has filed for a STAR Market IPO targeting ~$610-620M raise. Also announced: Cosmos 3 world model, Nvidia deepening its physical AI push, and Cosmos/Isaac integration for humanoid reasoning.
BYD's EVP Stella Li confirmed the company is developing humanoid robots and plans to sell them through its dealership network globally. 150 prototypes in BYD factories now. Target: 20,000 units deployed internally this year. A new Xi'an industrial park is designed for 50K/year production capacity. BYD frames it as an open robotics platform, combining automotive AI foundations with robot motion control.
The Musk irony is thick: BYD already outsells Tesla in cars, and now it's entering humanoid robots with actual manufacturing scale while Optimus remains a demo project.
SpaceX's IPO roadshow is live. $135/share, $1.75T valuation, $75B raise -- the largest IPO in history (3x Saudi Aramco). Reuters reports demand is already 2x oversubscribed. Fixed price, no bookbuilding -- Musk upending Wall Street tradition. Pricing June 11, trading June 12 as SPCX on Nasdaq. Goldman's CEO David Solomon personally signing off on allocations.
Relevant because: SpaceX, Anthropic, and OpenAI are all heading to public markets in 2026. This is the benchmark IPO that sets the tone for AI company valuations.
"Microsoft just banned Claude Code for 100,000 engineers" -- a viral LinkedIn post with 45+ comments breaking down the economics: token costs exceeding human costs, Uber blowing its budget, NVIDIA VP admitting compute costs surpass employee costs.
Anthropic's "When AI Builds Itself" report is getting wide coverage today. The 80%-Claude-authored-code stat and the coordinated-pause proposal are both generating strong reactions. Scientific American ran a prominent piece, and crypto/tech media picked it up fast.
A LinkedIn post from Qualixar gaining traction. Walks through Opus 4.8 Dynamic Workflows, Cursor Agent Mode's 68% SWE-Bench score (meaning 1 in 3 outputs is wrong), and Anthropic's shift to metered billing making unverified agents a direct financial risk.
Multiple outlets previewing WWDC (June 8). Apple rebuilding Siri as a full chatbot (standalone app, Dynamic Island integration, "Search or Ask" replacing Siri Suggestions). Tim Cook's final keynote. Siri still labeled "beta." Potential waitlist gating.
Thread idea: "Three things happened this week that signal the real robotics platform war has begun." Walk through (1) Nvidia + Unitree = the "Wintel of robotics" play at GTC Taipei, (2) BYD entering with 20K units in its own factories and dealership distribution, (3) Unitree's IPO filing valuing the company at the intersection of hardware scale and AI brains. The insight: the competition is shifting from motion control (solved) to AI "brains" -- and whoever builds the developer platform (Nvidia's Isaac GR00T) wins the same way CUDA won deep learning. Tesla Optimus is the vision, but BYD and Unitree are shipping units.
Tailored to Raunaq's lane: robotics analysis + semiconductor/AI investing (Nvidia's physical AI strategy directly connects to his TSM/ASML thesis).
Post idea: "Microsoft just banned Claude Code for 100K engineers. The math is simple: token prices drop 90%, but agentic workflows burn 5-30x more tokens than chatbots. Net effect? Bills go up. This is the tokenmaxxing paradox." Connect to the Anthropic metered billing shift (June 15), Uber's budget blowout, and Nvidia VP's admission. The punchline: the companies building AI say it saves money; the companies using AI are finding it costs more than humans. This is the infrastructure story nobody's pricing correctly.
Tailored to Raunaq's lane: AI agent infra founder who's directly dealt with token economics building on Claude. Credible voice on this, especially given the tokenmaxxing research he's already done.
Post idea: "SpaceX ($1.75T, June 12). Anthropic ($965B, fall). OpenAI (TBD). In 6 months, the three most consequential AI-adjacent companies will be public. Here's what each one tells you about where AI value accretes." SpaceX = infrastructure (Starlink + AI compute). Anthropic = models + developer tools (Claude Code driving $47B ARR). OpenAI = consumer + platform. The real question: which layer captures the most value long-term? This is semiconductor-to-application stack analysis, which is Raunaq's natural territory.
Dare Obasanjo (@carnage4life) is the most active AI voice on Threads right now -- 6+ posts in the last 24 hours, consistently high engagement (60-374 likes per post, 30+ comments on top threads). He's covering the AI agent economics story, Anthropic's RSI report, and enterprise AI adoption patterns. His audience overlaps significantly with who Raunaq should be reaching.
Elvis Saravia (@omarsar0, DAIR.AI founder) posted yesterday about AI research developments. His audience is technical and research-oriented. Consistent engagement from the ML research community.