not much happened today

**Anthropic** introduces durable agents and MCP tasks for long-running workflows, with practical engineering patterns and integrations like Prefect. **Booking.com** deploys a large-scale agent system improving customer satisfaction using LangGraph, Kubernetes, GPT-4 Mini, and Weaviate. **Perplexity** rolls out user-level memory and virtual try-on features. **Claude Opus 4.5** leads on LisanBench and Code Arena WebDev benchmarks with mixed community feedback on its "thinking" and "non-thinking" modes, while improving cost-efficiency and UX with batch APIs and context compaction. Research on multi-agent systems shows **LatentMAS** reduces communication tokens by 70-84% and improves accuracy using Qwen3 models, and reasoning trace distillation achieves significant token reduction with maintained accuracy, highlighting the importance of reasoning trace style.