CrewAI 1.15.3: Enhanced Control Over AI Agent Runs

CrewAI version 1.15.3 introduces new control features for AI agents. This update adds interception points within agent runs, particularly around model and tool operations. Tool-result caching is now opt-in, meaning data reuse requires a deliberate decision. The release also exposes per-call usage in kickoff results and fixes issues with hooks and native tools. These changes aim to give operators more control over AI agent actions. The update is described as a "control-surface release" rather than a new model or benchmark. It allows developers to observe or influence an agent's work while it's still running. For example, before-LLM hooks can inspect or modify messages, and before-tool hooks can validate inputs or refuse execution. This is important because it allows operators to stop, inspect, modify, and reproduce what happens during an agent's process, rather than just seeing the final outcome.