Aller au contenu
AI.info

events

Build Self-Improving AI Agents: Turn Failing Traces Into Higher pass@k

Webinaire gratuit consacré à l’analyse des traces d’agents en échec, au regroupement des modes d’échec et à la mise à jour automatique des instructions, de la mémoire et des outils pour améliorer…

Build Self-Improving AI Agents: Turn Failing Traces Into Higher pass@k | Événements de l’IA

En anglais

Starts
2026-12-02 19:00 UTC
Organizer
Data Science Dojo – Seattle
Location
Online event

Part five of a coding agent series hosted by Data Science Dojo, a free live online session with Kwasi Ankomah, Lead AI Architect at SambaNova Systems. The talk covers reading failing agent traces, grouping individual failures into recurring failure modes, and automatically updating an agent's instructions, memory and tools from those patterns using off-the-shelf libraries. It closes with A/B testing the improved agent against the baseline to measure gains in pass@k, without fine-tuning. Attendees should already be working with agents and evaluation.

Event website