<p><strong>TLDR:</strong> SpaceX secured an option to acquire Cursor for $60 billion, pairing Cursor’s distribution with xAI’s Colossus compute to build what they’re calling the world’s best coding AI. Anthropic had a rough day: it briefly removed Claude Code from the $20 Pro plan, got called out publicly, and reversed course within hours. OpenAI shipped gpt-image-2, which is genuinely impressive and now sits at the top of image generation leaderboards. GitHub paused new Copilot sign-ups because agentic workflows are destroying the economics of flat-rate pricing.</p>
<hr>
<h2>SpaceX and Cursor announce partnership, with a $60B acquisition option</h2>
<p>SpaceX has secured an option to buy Cursor outright for $60 billion before end of year, or pay $10 billion for the joint AI coding work if the acquisition doesn’t happen. The deal pairs Cursor’s product and its distribution to software engineers with SpaceX’s Colossus supercomputer — roughly a million H100 equivalents — to train new models. Cursor’s valuation went from $2.5 billion to $29.3 billion in under two years, so the $60 billion figure is essentially a bet on xAI’s ability to close the gap with Anthropic and OpenAI on coding benchmarks before the option expires. The more immediate question is what happens to Cursor’s multi-model flexibility: Anthropic cut off Windsurf’s model access when Windsurf was in acquisition talks with OpenAI, and the same dynamic could play out here in reverse.</p>
<h2>Anthropic’s Claude Code pricing stumble</h2>
<p>Anthropic quietly updated its pricing page to remove Claude Code from the $20 Pro plan, making it exclusive to the $100+ Max tier. The change went live without any announcement, got spotted almost immediately, and blew up on Reddit, Hacker News, and Twitter. An Anthropic employee described it as an A/B test affecting roughly 2% of new sign-ups. The reversal happened within hours, but the trust damage from the lack of transparency is harder to undo. The underlying pressure is real: Uber reportedly exhausted its entire 2026 AI budget because Claude Code usage ramped faster than expected, and Anthropic’s annualized revenue has gone from $9 billion to $30 billion. The economics of flat-rate plans are breaking down across the board.</p>
<h2>GitHub pauses new Copilot sign-ups</h2>
<p>GitHub stopped accepting new sign-ups for Copilot Pro, Pro+, and Student plans on April 20. The stated reason is straightforward: agentic coding workflows consume far more compute than the original pricing model was built to handle. Weekly infrastructure costs have reportedly doubled since January. Beyond the sign-up pause, GitHub is tightening limits across the board — Opus models are now Pro+ only, and Pro+ gets more than 5x the limits of Pro at $39/month. The practical takeaway is that AI coding tools should probably be evaluated as metered infrastructure now, not unlimited productivity layers.</p>
<h2>OpenAI ships gpt-image-2</h2>
<p>OpenAI released gpt-image-2, now available via API and in Codex. The headline improvement is text rendering — older diffusion models were notoriously bad at spelling, and this one handles dense text accurately including non-Latin scripts. It runs in two modes: Instant for fast output and Thinking for deliberate multi-step generation. It’s taken the top spot across image generation leaderboards. DALL-E 2 and 3 are being retired on May 12, making this a direct replacement. Codex has also hit 4 million active users, adding over 1 million in less than two weeks.</p>
<hr>
<h2>Also notable</h2>
<ul>
<li><strong>Amazon deepens Anthropic investment:</strong> Amazon is putting up to $25 billion more into Anthropic ($5B immediate, up to $20B tied to milestones), with Anthropic committing to over $100 billion in AWS spending over the next decade and up to 5GW of Trainium compute capacity.</li>
<li><strong>Kimi K2.6 in Windsurf:</strong> Kimi K2.6 is now free in Windsurf for Pro, Teams, and Max users for two weeks — multiple engineers report it’s competitive with Claude on coding tasks at a fraction of the price.</li>
<li><strong>Claude Design launches:</strong> Anthropic released Claude Design, a UI prototyping tool powered by Opus 4.7 that converts Figma designs into interactive prototypes and production-ready UIs, built in partnership with Canva. Early reactions are positive but note it’s slow and rate-limited separately from regular Claude.</li>
<li><strong>Google open-sources <a href="http://DESIGN.md" target="_blank" rel="noopener nofollow">DESIGN.md</a>:</strong> Google’s Stitch team published the <a href="http://DESIGN.md" target="_blank" rel="noopener nofollow">DESIGN.md</a> spec on GitHub — a file format for describing design systems to AI agents that combines human-readable reasoning with structured design tokens. The contrast with Anthropic’s rate-limited Claude Design product has not gone unnoticed.</li>
<li><strong>Anthropic’s Mythos cybersecurity tool leaked:</strong> An unauthorized group reportedly accessed Mythos, Anthropic’s exclusive AI cybersecurity model, through a third-party vendor environment by guessing the URL from known Anthropic URL patterns. Anthropic says its own systems were not impacted.</li>
<li><strong>Claude Code security bug:</strong> A hard-coded <code>MAX_SUBCOMMANDS_FOR_SECURITY_CHECK = 50</code> in Claude Code’s <code>bashPermissions.ts</code> means all deny rules silently stop being enforced after 50 subcommands in a session.</li>
<li><strong>Codex hits 4M users:</strong> OpenAI’s Codex has reached 4 million active users, up roughly 20x year-to-date.</li>
<li><strong>Qwen 3.6 35B fine-tune trending:</strong> A fine-tune of Qwen 3.6 35B on Claude Opus 4.6 distilled reasoning traces hit number one trending on Hugging Face, aiming to combine Opus-level reasoning with 35B inference speed.</li>
<li><strong>Google Deep Research Max:</strong> Google launched updated Deep Research and Deep Research Max in the Gemini API with MCP support and native chart generation — API only, not yet in the Gemini app.</li>
<li><strong>Roo Code shutting down:</strong> Roo Code is archiving its repo on May 15 after 3 million installs. Kilo, which started as a Roo fork, is continuing development and has rebuilt the extension on the OpenCode server.</li>
<li><strong>Token costs running over budget:</strong> Multiple reports from inside tech companies indicate AI token spend is coming in significantly higher than forecasted, with some companies spending up to 10% of headcount costs on LLM tokens.</li>
<li><strong>Hermes with Gemma 4 or Qwen 3.5 for local use:</strong> Several engineers are reporting this combination as the best local setup available before spending on cloud API credits.</li>
</ul>


AI coding assistants, code generation tools, and vibe coding workflows.

Agentic Digest

A roundup of major AI coding tool developments: SpaceX secured a $60B acquisition option for Cursor, pairing it with xAI's Colossus supercomputer to build coding AI. Anthropic briefly removed Claude Code from its $20 Pro plan without announcement, faced public backlash, and reversed within hours — highlighting the economic strain of flat-rate AI pricing. GitHub paused new Copilot sign-ups as agentic workflows doubled weekly infrastructure costs since January. OpenAI shipped gpt-image-2 with strong text rendering, retiring DALL-E 2 and 3 on May 12. Additional notes cover Amazon's deepened Anthropic investment, a Claude Code security bug, Roo Code shutting down, and widespread reports of AI token spend exceeding forecasts.

SpaceX bets $60B on Cursor, Anthropic's Claude Code pricing stumble