이미지: AWS ML Blog
Summary
- Mobileye has deployed an AI support agent built on Amazon Bedrock AgentCore
- The system automated repetitive status-check tasks that made up 66% of tickets, cutting response times by 90%
- The agent uses Anthropic's Claude models and MCP to access internal system APIs in real time
- 적용 기업
- Mobileye
- 사용 플랫폼
- Amazon Bedrock AgentCore
- 응답 시간 단축
- 90%
- 목표 정확도
- 95% 이상
- 기존 반복 티켓 비중
- 66%
- 기반 모델
- Anthropic Claude (Bedrock 경유)
Mobileye announced it has introduced an AI support agent based on Amazon Bedrock AgentCore to resolve support bottlenecks in its data collection and processing pipeline.
Repetitive Tasks Were the Bottleneck
Mobileye's data collection pipeline processes thousands of driving record sessions every day, and 66% of support tickets consisted of simple status-check requests that engineers had to verify manually, clicking through multiple backend systems up to 15 times. Script- or rule-based automation reportedly struggled to handle the wide variety of real-world inquiry patterns.
Results After Switching to an AI Agent
Mobileye set targets of 95% accuracy and response times under two minutes, ran a proof of concept, and then moved to full deployment. The agent uses Anthropic's Claude models through an internal LLM gateway, and leverages the Model Context Protocol (MCP) to access driving data processing platform APIs in real time — handling session status checks, log review, and diagnostic data collection.
As a result, response times dropped by 90%, and the accuracy target was exceeded. Mobileye stated it had built an "agent capable of understanding context and adapting to diverse inquiry patterns."
Following these results, Mobileye reportedly expanded AgentCore into a self-service platform that allows multiple internal teams to deploy their own AI agents.
