The June 4 SDK changelog adds function definitions directly on Agent.create(), so a single-function agent no longer needs a separate MCP server. We broke down when to use the new custom tools vs. when a full MCP server still wins. Read our decision guide. Try Cursor.

The Financial Times reported on June 7, citing more than a dozen current and former employees, that OpenAI will bundle Codex, AI agents, and partner integrations (Canva, Booking.com) into a single superapp shell. Unconfirmed by OpenAI. Codex already has 5 million weekly active users as of June 3. If it ships, every standalone agent builder competes against a default.

The keynote starts at 10 AM PT today. Before the event, Bloomberg reported (confirmed by Apple's WWDC invitation design) that Apple has partnered with Google to power the next-generation Siri with Gemini AI models. A standalone Siri app (comparable to ChatGPT and Claude) is also reported. Watch for what Apple actually announces; the Gemini deal is the one piece confirmed pre-keynote.

Anthropic announced on June 2 that its autonomous vulnerability-scanning program (powered by Claude Mythos Preview) is adding 150 organizations across power, water, healthcare, communications, and hardware. The initial cohort of about 50 organizations already found more than 10,000 high- or critical-severity security flaws. If your codebase touches infrastructure, that attack surface is about to get scanned at scale.

Polling a page every five minutes burns tokens and misses nothing most of the time. Firecrawl's /monitor endpoint pushes a structured diff only when content changes, cutting scrape spend by up to 90% per Firecrawl's May 2026 launch post. We walked through the complete n8n and Make builds side by side, with the per-run cost math that decides which platform fits. Read the Firecrawl /monitor guide. Try Firecrawl. Try Make.

OpenAI's Dreaming V3 processes your entire chat archive in the background, no manual setup. Claude Projects requires explicit curation; Gemini Past Chats stays session-scoped. We compared all three, with a pick per persona: power user, privacy-first operator, Google Workspace team. Read the memory comparison.

Usage-based billing means your AI spend scales with how you prompt. We broke down seven specific tactics (context pruning, model-tier routing, prompt caching, and four more) with the per-token math. Read the cost-cutting guide.

The TrustFall vulnerability lets a malicious repo execute arbitrary code through Claude Code, Cursor CLI, Gemini CLI, and GitHub Copilot the moment you approve folder trust. We mapped the mechanism, covered Anthropic's patch, and built the hardening checklist for your dev machine. Read the TrustFall hardening guide.