Microsoft killed Claude Code internally — the same week Anthropic doubled its limits

Hey — Two stories collided this week and you really only need both of them to understand where the AI coding market is right now. Microsoft is pulling Claude Code licenses from its own engineers and pushing them to GitHub Copilot CLI. Three days earlier, Anthropic *doubled* Claude Code's rate limits overnight after a SpaceX deal brought 300 megawatts of new compute online — the equivalent of 220,000+ Nvidia GPUs in under a month. The flat-rate AI coding era is over, the platform politics are getting ugly, and Google I/O is in 48 hours. ## This Week in AI Coding - **Microsoft yanks Claude Code from its own engineers.** Internal memo first reported by [Windows Central](https://www.windowscentral.com/microsoft/microsoft-cancels-claude-code-licenses-shifting-developers-to-github-copilot-cli-a-move-likely-driven-by-financial-motives): MSFT is cancelling Claude Code seats company-wide and routing developers to GitHub Copilot CLI. Framed internally as a cost decision; reads externally as platform politics. If you needed proof that "best tool" loses to "tool we own" inside big tech, this is it. - **Anthropic doubled Claude Code rate limits overnight.** Same week, opposite direction. Per [SD Times' weekly roundup](https://sdtimes.com/ai/may-8-2026-ai-updates-from-the-past-week-coder-agents-launch-snyk-claude-partnership-opsera-cursor-partnership-and-more/), Anthropic spun up 300 MW of new compute via a SpaceX partnership and immediately handed it to Claude Code users. If you've been getting rate-limited on Opus 4.7, retry today. - **GitHub Copilot CLI shipped 5 releases in 6 days.** v1.0.44 fixed path completion flicker, v1.0.45 added a `/autopilot` slash command to toggle interactive vs. autonomous mode. GitHub is sprinting because they have to — and because they just paused new Copilot Pro/Pro+ signups ahead of a [usage-based billing switch June 1](https://www.windowscentral.com/microsoft/microsoft-cancels-claude-code-licenses-shifting-developers-to-github-copilot-cli-a-move-likely-driven-by-financial-motives). If you're on a grandfathered seat, hold it. - **Cursor is now inside Microsoft Teams.** Announced May 11. `@cursor` in any Teams channel delegates work to a Cursor agent or pulls context from your repo into the conversation. Cursor's Bugbot also flips to usage-based billing [June 8](https://medium.com/@chaos.architect25/the-best-ai-coding-tools-of-may-2026-cf2db2804a0f). Same direction as Copilot — the seat era is dying. - **GPT-5.5 Instant is now the default ChatGPT model.** Quietly swapped in [May 5](https://techcrunch.com/2026/05/05/openai-releases-gpt-5-5-instant-a-new-default-model-for-chatgpt/), replacing 5.3 Instant. Codex on iOS + Android is now rolling out to Free and Go tiers too. OpenAI's quiet move while everyone watched the platform fights. - **Google I/O is Monday — Gemini 4 expected.** [May 19-20 at Shoreline](https://io.google/2026/). Leaked specs say [2M-token context window, sub-300ms latency, 84.6% on ARC-AGI2](https://www.abhs.in/blog/google-io-2026-preview-gemini-4-android-17-developer-guide). The real story: Google's own agentic coding tool — their answer to Claude Code and Cursor — is reportedly part of the keynote. If that ships well, the 3-horse race becomes a 4-horse race overnight. - **xAI retired old Grok model IDs May 15.** Quiet but breaking: anything still hitting xAI API with pre-Grok-4.3 model names is dead. Worth a 30-second `grep` of your codebase. - **AI layoffs accelerated.** Meta announces 10% cuts (~8,000 roles) starting May 20. Coinbase cut 14% earlier in the month. Per [CNBC](https://www.cnbc.com/2026/04/24/20k-job-cuts-at-meta-microsoft-raise-concern-of-ai-labor-crisis-.html), new software engineering postings are down 15% YoY. The pattern is sharper now — implementation-speed roles are taking it hardest, domain-judgment roles are barely touched. ## What We Shipped - **6 new videos on [@endofcoding](https://www.youtube.com/@endofcoding)** this week — total now 122 videos live. Daily Sun-Thu cadence holding. EP115 ("vibe coding patterns that break in production") is in production for the next drop. - **SPARit had a +570% engagement breakout** after Tuesday's content push and picked up its first real ideator signup (Yael, May 14). The YouTube engine for SPARit is back online; first short went live this week. - **LLMHire crossed 5,793 indexed AI jobs.** Live data, filterable by role, salary, remote/onsite, and skills required. Free to search at [llmhire.com](https://llmhire.com) — and that's the dataset I keep pulling from when these layoff stories drop. - **Lemon Squeezy live across CyberOS, the Vibe Coding Ebook, and LLMHire** continues humming. Test payment flow re-verified this week. ## Tool of the Week: GitHub Copilot CLI `/autopilot` Bury the lede: the most interesting *feature* shipped this week wasn't from Anthropic or Cursor — it was Copilot CLI's `/autopilot` toggle. You're in an interactive shell session, you type `/autopilot`, and the agent flips into headless mode and just *keeps going* on the multi-step task without asking. Then `/autopilot` again to drop back into the conversation. It's the cleanest implementation of "trust mode" I've seen — much better than the "yes to all" hacks everyone else ships. If you're already on Copilot Pro and you've been wary of long-running agent runs, give this 20 minutes. Combined with the June 1 billing change, it's a strong signal that even GitHub thinks the future is "set it and walk away." ## Quick Links - [Microsoft cancels Claude Code licenses — full memo coverage](https://www.windowscentral.com/microsoft/microsoft-cancels-claude-code-licenses-shifting-developers-to-github-copilot-cli-a-move-likely-driven-by-financial-motives) - [Anthropic + SpaceX: 300MW of new compute](https://sdtimes.com/ai/may-8-2026-ai-updates-from-the-past-week-coder-agents-launch-snyk-claude-partnership-opsera-cursor-partnership-and-more/) - [Google I/O 2026 — livestream Monday 10am PT](https://io.google/2026/) - [The Best LLMs for Agentic Coding in 2026 — real-world benchmark](https://dev.to/danishashko/the-best-llms-for-agentic-coding-in-2026-real-world-not-just-benchmarks-96n) ## What I'm Building With This Week Every newsletter, the same stack runs in the background: Claude Code for the heavy lifting, Cursor for the IDE work, GitHub Copilot CLI as a second opinion. This week the platform politics finally cracked through — I'm now running parallel agent sessions on Opus 4.7 (newly un-rate-limited) to stress-test which tasks actually benefit from the bigger context window vs. which just burn tokens. The Vibe Coding Ebook covers exactly this multi-agent stack approach — including the new patterns from the post-Karpathy "agentic engineering" frame. — **Built with AI agents in Tel Aviv.** [endofcoding.com](https://endofcoding.com) · [vibecodingebook.com](https://vibecodingebook.com) · [llmhire.com](https://llmhire.com) · [cyberos.dev](https://cyberos.dev) · [YouTube @endofcoding](https://www.youtube.com/@endofcoding) If a friend forwarded this, [subscribe here](https://buttondown.com/endofcoding).