59The agent works in the pilot and falls apart somewhere between ten and fifty concurrent runs. Diagnose it before you add capacity.▼hardNewOpenAIAnthropicDatabricks4 replies◆ premiumEveryone sizes the pool on average run length, and average is the one statistic an agent fleet does not have. Simulated, the p99 sat at 160 seconds whether the system was half loaded or nearly saturated, which means capacity was never the problem.Open full answer →
64Your agent abstains on 8% of cases and every one becomes a human's problem. Design the escalation system.▼hardNewSierraDecagonHarvey◆ premiumThe abstain path is designed last and decides whether the deployment is adopted. Routing to the right person, an SLA the queue can actually meet, and what happens to the case the agent already half-processed before it gave up.Open full answer →
