Microsoft is canceling Claude Code licenses for thousands of engineers in its Experiences + Devices division (Windows, Office, Teams, Surface) by June 30. The backstory is damning for Copilot: Microsoft gave engineers Claude Code in Dec 2025 as a pilot, and they loved it so much they stopped using Copilot CLI entirely. Now Microsoft is forcing them back. Meanwhile, GitHub Copilot's new token-based billing (live June 1) has developers reporting 25x bill jumps overnight, with Claude Opus model multipliers going from 3x to 15x.
The Bridge Chronicle, Memeburn, WebProNews -- June 3-4
Broadcom lost ~$315B in market cap today after Q2 earnings. The numbers were solid: $22.2B revenue (+48%), AI semi revenue $10.8B (+143%), Q3 guide $29.4B (above estimates). But the market wanted them to raise the $100B FY2027 AI revenue target and they held it flat. "Classic case of very high expectations meeting a market that wanted perfection." MRVL -5.8%, AMD down, NVDA -0.8%. The Dow rallied 1.7% while Nasdaq dipped -- a sector rotation day.
Reuters, MarketWatch, Investopedia -- June 4
Anthropic has confidentially filed with the SEC, days after closing a $65B Series H at a $965B valuation (surpassing OpenAI's $852B). The IPO could raise $60B+. This sets up an unprecedented trio of potentially trillion-dollar tech IPOs: Anthropic, SpaceX ($1.75T target, June 12), and eventually OpenAI. Claude Code is explicitly cited as the growth engine that let Anthropic catch up on revenue. The quiet period implications alone are interesting -- Dario Amodei and Sam Altman will have to dial back public commentary.
Reuters, The Verge, Motley Fool -- June 2-4
Codex is becoming ChatGPT's agentic backbone. OpenAI announced Codex functionality coming to the ChatGPT app "in the next few weeks," plus six role-specific enterprise plugins (Sales via Pipedrive, Revenue via Outreach, Web via Wix, etc.), a new "Sites" feature for hosting interactive apps/dashboards, and an annotations system for editing documents, spreadsheets, and slides in-place. Altman's follow-up focus: reducing token costs. Partners include Vercel, Figma, Webflow, Replit, and Lovable.
9to5Mac, GlobeNewsWire, OpenAI Livestream -- June 3
Figure AI + Catalyst Brands: Figure signed its first retail-sector deal to deploy humanoids at Catalyst's Reno warehouse (JCPenney, Brooks Brothers, Eddie Bauer parent). Brookfield is the investor connecting both sides. Amazon Proteus: Next-gen warehouse robot now understands conversational prompts -- "You tell it what needs to be done. It figures out the priority, the route, the timing." Part of a $11.6B European fulfillment push. Also unveiled Vulcan (touch-sensing robot) and STARK (tote-handling). BMW: After Figure 02 logged 1,250 hours and contributed to 30,000+ X3 assemblies in Spartanburg, BMW is expanding humanoids to Leipzig with Hexagon Robotics AEON units.
Fox News, Reuters, Engadget, WebProNews -- June 4
This story is everywhere today. Developers are sharing hot takes on Claude Code vs Copilot CLI, and the GitHub Copilot billing shock is compounding the narrative. Multiple LinkedIn posts with 50-100+ comments on the topic.
OpenAI just launched a hosting product for Codex-generated interactive apps and dashboards. This is directly in html-docs territory. Dev Twitter is buzzing about what this means for Vercel, Replit, and the broader "AI builds you a website" stack.
Eduardo Ordax's post (73 comments) on an Anthropic engineer's talk about Claude Code orchestration patterns: CLAUDE.md burning 14% of context, automation workflows most users don't know exist, daily task pipelines running without the keyboard. The framing: "Stop using the chat window. Build the system."
Massive discussion around Anthropic's S-1 filing, $965B valuation, and what Claude Code means for revenue. Multiple threads dissecting whether Anthropic can sustain enterprise growth and how the quiet period will affect their aggressive marketing.
Thread idea for X. Today, in a single 24-hour window: Figure AI signed its first retail-sector deployment (JCPenney's parent), Amazon revealed a warehouse robot that takes orders in plain English, and BMW confirmed European expansion after 30,000 cars assembled with humanoid help. This isn't the "robots are coming" narrative anymore. The physical AI deployment curve just hit inflection. Tie it to the investment angle -- Zacks ran "Are humanoid robots the next trillion-dollar theme?" and Masayoshi Son said the next trillion-dollar company comes from physical AI. Your lane: you track both the AI infra side (semiconductors, agent software) and the physical deployment side. Few people cover both.
Thread idea for X. OpenAI just launched Codex Sites (hosting for agent-generated apps). Anthropic's Claude Code already generates full projects. Google's Gemini creates interactive artifacts. Every major AI lab is now building an output hosting layer -- but they're all building walled gardens. The pattern is familiar: the value isn't in the compute layer (commoditizing fast), it's in the persistent artifact layer where humans actually interact with what agents produce. HTML is the new markdown, and the output format war is just beginning. Directly positions html-docs without naming it -- let the replies surface the connection.
Short post or thread for X. Broadcom grew AI revenue 143% YoY, guided Q3 above estimates, and still lost 15% of its market cap because it didn't raise a target it set three months ago. This is the AI semi market in one chart: the bar for "good enough" keeps rising. Custom silicon demand is real and accelerating, but the stocks have priced in not just success but perfection. What does this mean for the next wave of NVDA, MRVL, and TSM earnings? The semi trade isn't broken -- but it requires conviction through these pullbacks.
@saboo_shubham_ (13K followers) shipped a collection of Generative UI Agents to the Awesome LLM Apps repo (113K+ stars on GitHub). The thesis: the traditional frontend is dead for agent-driven products; agents should generate their own UI dynamically. This is a Threads content opportunity -- the "generative UI" framing pairs naturally with html-docs as the hosting/rendering layer for agent-generated interfaces.
Posted June 3, 56 likes, climbing
@hanifmuh_ (163K followers) ranked the top 5 cost-efficient models for Hermes/OpenClaw agents: MiMo-V2.5, DeepSeek V4 Flash variants, MiMo-V2-Flash. 199 likes and climbing. The discussion underneath is about whether agent cost optimization matters more than raw capability -- a topic where Raunaq can weigh in from the html-docs angle (agents running at scale need cheap inference AND a persistent output layer to avoid re-generation).
Posted June 3, 199 likes