The July 17 update turns /fork into a true background session visible in claude agents, adds per-session caps on WebSearch (default 200) and subagent spawns (default 200) to stop runaway loops, and improves cold-attach behavior for background sessions.
Why it matters: Built-in loop guards and stable background session management make long-running agentic workflows more predictable — a direct reliability improvement for teams running autonomous tasks at scale.
6sense now exposes account intent, buying-stage predictions, and campaign data directly inside Claude, ChatGPT, Writer, and Agentforce via MCP with no custom integration required; GA planned for August 2026.
Why it matters: MCP-native GTM intelligence removes custom integration overhead — account intent and predictive data become callable from any compatible AI agent stack, making it easier to build revenue-aware agents.
Google's rebuilt Gemini 3.5 Pro missed its July 17 target for the third time; engineering teams are still working through hallucination issues and Google is reportedly exploring a partial stopgap release rather than the full 2M-context model.
Why it matters: For teams evaluating Gemini 3.5 Pro for production use, don't plan around it until there's an official launch — the 2M-token context window and Deep Think layer remain unavailable while Google debugs reliability.