Meet the new Rovo Chat: One prompt, multiple steps, zero hand-holding

This post details the transition from a multi-agent routing system to a single-model reasoning engine ('Long Horizon') for Rovo Chat. The new engine uses a single model to hold full context across tools, enabling iterative reasoning (think, call tools, observe, reflect, repeat) up to 150 times per query. It employs adaptive reasoning effort based on query complexity. A tradeoff of increased latency for complex queries is mitigated by exposing reasoning traces. Offline evaluations show an 8.5% improvement in answer quality, a 23% improvement on Confluence tasks, and a 37% reduction in perceived latency due to real-time progress visibility. New capabilities include agentic tool use for cross-product workflows and long-running background tasks. The underlying stack is upgraded with Claude Opus 4.x class models and improved Teamwork Graph connectors.