A significant study of 157 AI agent deployments revealed that planning quality, not execution speed or model size, is the primary determinant of success. This observation has led to the development of "Orca-style" agents, which are hierarchical and prioritize planning. The research involved varying agent architecture, planning depth, and execution models across diverse use cases. Agents dedicating more tokens to planning achieved significantly higher task completion rates and fewer rollbacks.The economic principle behind this is that planning is inexpensive compared to the high cost of correcting mistakes made during execution. Orca-style agents separate planning from execution, with a strategic planner handling complex reasoning and specialized executors performing defined tasks. Specialists have "skill cards" describing their capabilities, and a shared memory layer maintains state.Key effective implementation patterns include recursive decomposition with validation gates, specialist routing, and stateful context propagation. Pitfalls to avoid are over-planning, excessive specialist fragmentation, silent replanning, and context window hoarding. Orca-style fleets are recommended for multi-step workflows and high-stakes operations.The cost model favors Orca-style architectures for complex tasks, showing reduced total token costs due to fewer retries and escalations. The future of AI agents will likely focus on enhancing planning capabilities with dedicated tools and libraries. The core insight is that thoughtful planning is the most crucial aspect of agentic work. Implementing Orca-style planning can begin with a single planner function that decomposes goals and validates each step.