June 19's news is dominated by the infrastructure layer catching up to agentic AI: Vercel shipped 'Eve' and new enterprise controls for running agents in production, Replit folded itself directly into Claude, and Okta partnered with Google Cloud to extend identity and zero-trust controls to AI agents and browsers. On the security side, researchers picked apart how Anthropic actually sandboxes its hosted Claude agents, and a widely-shared thread argues engineers should treat autonomous agents as untrusted third parties by default. Data infrastructure also got attention, with pushes toward self-hosted 'sovereign' AI stacks and entity resolution as a missing foundation for AI-ready data. Meanwhile ChatGPT's market share reportedly slipped below 50% for the first time as users spread out across Gemini and other assistants, and SpaceX was reported to be making a huge play for AI coding startup Cursor.

• Replit now runs as a tool inside Claude, blurring chat and IDE
• Vercel launches Eve plus enterprise controls for agentic infra
• Okta and Google Cloud extend identity/zero-trust to AI agents
• Researchers reverse-engineer Claude's hosted agent sandbox boundaries
• ChatGPT market share reportedly dips below 50% for first time