What matters in AI.

Subscribe

Agents

Ankur Duggal found bad output in 44% of the runs of an agent

The traces of the agent showed the fault, because they show each tool call and each document that it used.

44% of the agent's interactions ended in a bad summary A grid of 100 squares, 44 filled red, shows that 44 of 100 interactions of a financial agent ended in a bad summary, seen in traces. A few sample answers had looked fine. Financial agent, Arize AI demo 44% of agent interactions ended in a bad summary SEEN IN TRACES 44 OF 100 Few sample answers: looked fine
Ankur Duggal's financial agent seemed fine in a few sample answers, but its trace data showed 44% of interactions ended in a bad summary.

Sources

  1. Ankur Duggal Says 44% of an AI Agent's Outputs Were Bad Until He Looked at the TracesBigGo Finance
AI MATTER · NEWS · AI MATTER · NEWS ·11 OCT2026

Posted

More in Agents

All Agents news