Eduardo urges retry and refusal metrics

- AI engineer Eduardo argued observability must go beyond latency and cost to track retry rates and spikes in model refusals, posting on Apr 26. (x.com) - He specifically called for observability signals that expose escalating retry loops and sudden refusal spikes so teams can isolate model or tool failures fast. (x.com) - This advice matched other posts pushing hallucination, drift and output-quality metrics over raw throughput in agent contexts. (x.com)