Long Horizon: How Atlassian Built a Reasoning Engine for Complex AI Tasks

Long Horizon is a reasoning architecture replacing the previous hybrid orchestrator model, using one LLM with unified context, flattened tool surfaces, progressive disclosure, adaptive reasoning depth, context compaction, and child instance spawning for parallel research. Results show 8.5% improved accuracy and 23% better task completion versus production systems, with a 37% perceived latency reduction through streaming updates.