A team of developers built rudel.ai after struggling to understand their own Claude Code usage patterns. With no visibility into which sessions succeeded or why some got abandoned, they created an analytics layer that collected data from 1,573 real sessions totaling 15 million tokens and 270,000 interactions.

Their analysis uncovered surprising patterns in AI agent behavior. Skills were only used in 4% of sessions, while 26% of sessions ended within the first 60 seconds. Success rates varied dramatically by task type, with documentation work scoring highest and refactoring tasks performing worst. The team also identified error cascade patterns that could predict abandonment within the first two minutes.

These findings point to a critical gap in AI development tools: there's no established benchmark for what constitutes effective agentic session performance. The team is now building such metrics to help developers understand and improve their AI coding workflows. rudel.ai is available as a free, open-source tool that provides dashboards for token usage, session duration, and activity patterns, giving developers the insights they previously lacked.