Strong momentum around agent orchestration and persistent context as devs tackle multi-session workflows. Apple Core AI models land on iOS with solid performance data whilst established platforms like aisuite see renewed interest.

Model Context Protocol (MCP) is Claude's standard for sharing context between AI applications and external data sources. It appears in tools like Omnigent for agent coordination, ClariLayer for persistent data context, and the Terraform MCP Server for infrastructure management. It lets devs build agents that remember schemas and maintain context across sessions without rebuilding knowledge each time.

Context compaction intelligently summarises long conversation histories to fit within model token limits whilst preserving essential information. The technique identifies which messages contain key decisions, facts, or context and compresses redundant exchanges into concise summaries. Tools like fable-mode use staged planning to manage this process systematically. It's crucial for long-running agents that would otherwise lose important context or hit expensive token limits.