<p><strong>TLDR:</strong> SpaceX dropped its S-1 today targeting a $1.75 trillion valuation — the largest IPO in history if it prices — with Cursor acquisition plans and a bleeding xAI division buried in the filing. OpenAI separately claimed its reasoning model disproved an Erdős conjecture open since 1946, and the math community is cautiously calling it real. Google I/O continued generating noise across Gemini, Antigravity, and Android tooling. ClickUp cut 22% of staff while announcing $1M salaries for whoever survives.</p>
<hr>
<h2>SpaceX S-1: Starlink profits, xAI losses, and a $60B Cursor deal</h2>
<p>SpaceX filed for a Nasdaq IPO targeting $1.75 trillion — Starlink is the only profitable segment, generating $11B of $18.7B in 2025 revenue with a $1.19B operating profit in Q1 2026. The xAI division lost $6.4B in 2025 on $3.2B in revenue, and burned $2.47B in Q1 2026 alone. The filing also discloses plans to acquire Cursor for $60B roughly 30 days post-IPO — Cursor has over one million paying customers and $1B in annualized revenue, and would fold into the xAI empire to compete directly with GitHub Copilot and Claude Code. Musk retains 85.1% voting control through a dual-class structure.</p>
<h2>OpenAI claims AI disproved an Erdős conjecture</h2>
<p>OpenAI says an internal reasoning model disproved the planar unit distance conjecture, a geometry problem Paul Erdős posed in 1946. The model found a new family of point configurations that outperforms the grid-based solutions mathematicians had assumed were optimal for nearly 80 years. Mathematicians Noga Alon, Melanie Wood, and Thomas Bloom have all weighed in positively — Bloom called it elegant and intricate. This is meaningfully different from OpenAI’s earlier embarrassing episode where GPT-5 was credited with solving problems that already had solutions.</p>
<h2>ClickUp cuts 22% of staff, introduces $1M salaries</h2>
<p>ClickUp laid off 22% of its workforce as CEO Zeb Evans restructures around a model where AI agents outnumber employees 3:1. The company now runs roughly 3,000 internal AI agents and is introducing $1M salary bands for staff who demonstrate what Evans calls “100x impact.” The framing is that top engineers no longer write code but direct agents, and roles that haven’t adapted are structurally obsolete. It’s a template other companies are watching closely.</p>
<h2>GitLab 19.0 ships with agent platform expansion and usage-based billing</h2>
<p>GitLab 19.0 expands the Duo Agent Platform across the full software lifecycle, adds generally available SBOM-based dependency scanning for Maven, Gradle, and Python, and brings in Claude Opus 4.7, Gemini, Devstral 2, and GLM-5.1 model support. The billing model shifts to GitLab Credits — one dollar per credit, with Premium users getting 12 and Ultimate users 24 per month. Infrastructure changes include Valkey replacing Redis and PostgreSQL 17 as the new minimum.</p>
<hr>
<h2>Also notable</h2>
<ul>
<li><strong>Anthropic expanding to Colossus 2:</strong> Anthropic announced it’s scaling up GB200 capacity in SpaceX’s Colossus 2 data center throughout June — an interesting infrastructure dependency given the SpaceX IPO filing.</li>
<li><strong>Qwen 3.7 Max released:</strong> Alibaba’s new proprietary agentic coding model scores 56.6 on the Artificial Analysis Intelligence Index, outperforming Gemini 3.5 Flash and Kimi K2.6 on that benchmark.</li>
<li><strong>Modal raises $355M Series C at $4.65B:</strong> The GPU cloud platform reports fivefold growth since its Series B and over $300M in annualized revenue, with over 1 billion sandboxes launched to date.</li>
<li><strong>Codex Thursday updates:</strong> OpenAI shipped Appshots (add context from any Mac app directly to prompts), locked-screen Mac operation, a new annotation editor, and shareable plugins. OpenAI is also offering 2 free months of Codex to developers switching from competing tools.</li>
<li><strong>Devin gets a Windows VM:</strong> Cognition’s AI software agent now supports Windows virtual machines, demonstrated with a Minesweeper run.</li>
<li><strong>Grok Build 0.1 released:</strong> xAI quietly shipped a coding-focused model with a 256K context window, no output limits, priced at $1/M input and $2/M output tokens, positioned for long-horizon agentic tasks.</li>
<li><strong>Microsoft open-sources Rampart and Clarity:</strong> Two safety tools for agentic AI — Rampart runs continuous red-team testing in CI/CD, Clarity helps engineers validate design assumptions and trust boundaries earlier in the workflow.</li>
<li><strong>v0 adds PR conflict resolution:</strong> Vercel’s v0 now has a “Fix PR Conflicts” feature where an agent merges the base branch and resolves conflicts automatically.</li>
<li><strong>Gemini CLI adds subagents:</strong> Specialized expert agents with isolated context windows, their own tools and MCP servers, definable via Markdown files with YAML frontmatter.</li>
<li><strong>GitHub Copilot for Eclipse goes MIT:</strong> Full codebase open-sourced, exposing implementation details for completions, agent mode, MCP integration, and BYOK.</li>
<li><strong>Hark raises $700M Series A at $6B valuation:</strong> Brett Adcock’s new AI startup is building a “universal” agentic personal assistant with dedicated hardware, though almost no product details have been shared.</li>
<li><strong>Manus eyes $1B raise after China blocks Meta acquisition:</strong> Beijing’s NDRC ordered Meta’s $2B+ acquisition of Manus unwound, citing AI technology outflow concerns — a significant precedent showing Singapore redomicile no longer neutralizes Chinese investment review risk.</li>
<li><strong>Claude Mythos audited Symfony:</strong> Anthropic’s unreleased model found 19 confirmed vulnerabilities with zero false positives across Symfony and Twig; all 19 have been patched. A separate audit of curl found one low-severity issue from five submissions, three of which were false positives.</li>
<li><strong>Google Chrome silently installs 4GB AI model:</strong> Chrome downloads a weights.bin file for Gemini Nano features without user permission, re-downloads it after deletion, and the only way to stop it is disabling AI features via flags or uninstalling Chrome.</li>
<li><strong>Anthropic opening Milan office:</strong> EMEA is Anthropic’s fastest-growing region with run-rate revenue up ~9x year-on-year; the Milan launch is part of a broader Italy strategy that includes a Vatican AI ethics event appearance.</li>
<li><strong>GitHub Copilot moves to usage-based billing June 1:</strong> Code completions stay free, but chat and agent tasks shift to token-based AI Credits — power users running daily agent tasks could see costs jump 3-10x over current flat-rate pricing.</li>
</ul>


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

Agentic Digest

A packed week in tech: SpaceX filed for a $1.75T Nasdaq IPO with Starlink as its only profitable segment, plans to acquire Cursor for $60B, and xAI losing $6.4B in 2025. OpenAI claims its reasoning model disproved a 1946 Erdős geometry conjecture, with mathematicians cautiously validating the result. ClickUp cut 22% of staff while introducing $1M salary bands for engineers who can direct AI agents at scale. GitLab 19.0 shipped with agent platform expansion, new model support, and a shift to credits-based billing. Additional notable items include Anthropic scaling on Colossus 2, Qwen 3.7 Max release, Modal's $355M raise, GitHub Copilot moving to usage-based billing, and Google Chrome silently installing a 4GB Gemini Nano model.

SpaceX files for IPO at $1.75T, OpenAI claims first AI math breakthrough

<p>These digests are great but they sometimes hold the same information a previous digest already held, which is kinda annoying.</p>


<p>“users (…) could see costs jump 3-10x over current flat-rate pricing”
surprising :/</p>
