303K Tool Calls Later: 96% of Claude Code Errors Aren't the Model's Fault
Nitjsefnie · reddit · 2026-11-17
To test the claim that Claude Code is 'error after error', the author analyzed 303,205 tool calls across 1,618 Opus 5.5 sessions: only 5.5% errored, and of those, 53% were Bash commands legitimately exiting non-zero and 43% were the author's own PreToolUse hooks rejecting calls. Actual tool failures were just 182 calls — 0.06%. Error rate mostly measures your setup, not the model. They also released claudit, a self-hosted MIT dashboard for Claude Code usage, error-rate, cost, cache and latency analytics, built itself via multi-agent workflows.
More from coding & agent
- Neo4j dev advocate to walk Miami JUG through AI agents and MCP — mkheck · 2026-11-17
- Customer saves $9M+ by ditching API pricing and Devin for aggregated AI subscriptions — garrytan · 2026-11-17
- Miro's year-long experiment: making the canvas itself the AI prompt, not a chat sidebar — max__drake · 2026-11-17
- Workato CISO shares how AI agents triage ~80% of security alerts in production — adamse · 2026-11-17
- DEF CON talk breaks down 6 data-exfiltration vulnerabilities in AI agents — Kyrannio · 2026-11-17
- He wrote an MCP server for his label printer — now Claude prints labels for him — Ill-Watercress2519 · 2026-11-17