Mises à jour produit, notes de version, recherche et actualités de l'équipe Aera.
We swapped the LLM that picks an agent's memories for a classifier. Same coverage, higher precision, a third of the latency on unattended work. Here's where it failed.