AWS Launches AgentCore Optimization to Combat AI Agent Drift
SEATTLE — Amazon Web Services on Friday launched AgentCore Optimization in public preview, a new tool designed to address a growing problem in enterprise AI: the silent degradation of AI agent performance after deployment.
The service introduces what AWS calls an “agent performance loop” — a system that analyzes production traces from deployed AI agents, generates improvement recommendations, and validates those recommendations through batch evaluation and A/B testing before they reach production, according to a post on the AWS Machine Learning Blog.
The tool targets a pain point familiar to enterprises running AI agents at scale. Agents that perform well at launch tend to degrade over time as underlying models evolve, user behavior shifts, and prompts get reused in contexts they were never designed for, AWS said. In most engineering teams, detecting and addressing that quality drift remains a manual, reactive process.
AgentCore Optimization aims to close that gap by automating the detection-to-deployment cycle. The system monitors agent behavior through production traces, identifies patterns that indicate declining performance, and surfaces actionable recommendations — such as prompt adjustments or configuration changes — that teams can validate before shipping.
The preview launch positions AWS in the emerging agent observability and optimization market, where it faces competition from tools including Anthropic’s agent evaluation capabilities, LangSmith from LangChain, and a range of startups focused on AI agent monitoring.
The release comes as cloud providers expand their agent infrastructure offerings. AWS, Microsoft Azure, and Google Cloud have each broadened their agent platforms in 2026 as enterprise customers move AI agents from proof-of-concept to production workloads.
AgentCore Optimization is available immediately in public preview. AWS did not disclose pricing details for the service. The company said it is part of the broader Amazon Bedrock AgentCore suite, which provides infrastructure for building, deploying, and managing AI agents on AWS.
