Mobileye used Amazon Bedrock AgentCore to deploy an AI Support Agent for internal support operations, cutting ticket response times by 90% and exceeding a 95% accuracy target, according to an AWS Machine Learning Blog post published August 5, 2026. Mobileye, an autonomous driving company with more than 230 million EyeQ system-on-chips deployed across roughly 1,200 vehicle models worldwide, built the agent to reduce routine ticket-status work for skilled engineers.

Mobileye’s Data Collection Processing pipeline ingests thousands of drive-recording sessions daily, creating frequent status inquiries from engineers and data teams. Before the deployment, 66% of support tickets were routine status inquiries, and engineers had to move through 15 clicks across multiple backend systems to identify sessions, check visualization tools, validate outputs, review logs, and write responses.

Mobileye first ran a proof of concept targeting 95% accuracy in ticket classification and sub-2-minute response times. The agent used Anthropic Claude foundation models through Mobileye’s internal LLM Gateway, which provides governed and quota-managed access to foundation models on Amazon Bedrock. Model Context Protocol, or MCP (standardized tool connection for AI agents), let the agent query live drive-data processing APIs during inference, including session status, processing logs, and diagnostic information.

The agent handled more than basic ticket classification. For drive-recording session inquiries, it could confirm completed sessions with access details, surface specific errors with debugging recommendations and log links, or guide users through submission steps for missing requests, all in under two minutes without human intervention.

After the proof of concept, Mobileye chose Amazon Bedrock AgentCore for production because it offered serverless infrastructure, faster support resolution, built-in observability, support for multiple agentic frameworks, and hybrid architecture support. The hybrid design was required because Mobileye’s internal ticketing system ran on-premises and could not be accessed directly from AWS.

The production architecture split work across on-premises systems, AWS services, and Mobileye services. A Local Orchestrator extracted new tickets and posted completed responses back into the internal ticketing system. AgentCore Runtime ran the AI Support Agent with automatic scaling and single-API-call invocation. AgentCore Observability traced each agent interaction, including MCP tool calls, retrieved data, errors, session metrics, latency, token usage, and debugging traces.

AWS Secrets Manager stored, rotated, and retrieved credentials programmatically so teams did not handle secrets directly. Mobileye’s Data Pipeline MCP server gave the agent real-time access to live production data, including driving-session warming statuses, request progress, and processing errors. The AI LLM Gateway managed governed access to foundation models on Amazon Bedrock and other LLM providers.

In production, Mobileye reported a 98% overall success rate, above the original 95% target. Response time fell from hours to about 1 minute, a 90% improvement. The agent automated 66% of total ticket volume and processed 100+ tickets per month.

Mobileye later turned the approach into an internal managed service so developers could deploy production-grade AI agents without AWS expertise or cloud credentials. Teams provide agent code and specify AgentCore capabilities such as Memory, Browser Tool, Code Interpreter, Observability, and Gateway; Mobileye’s Cloud Infra team provisions AWS IAM Roles, Amazon S3 storage, Amazon CloudWatch monitoring, and Amazon Cognito authentication. Developers receive a preconfigured bedrock_agentcore.yaml file and deploy with a single command, agentcore deploy.

Mobileye used Amazon Bedrock AgentCore to deploy an AI Support Agent for internal support operations, cutting ticket response times by 90% and exceeding a 95% accuracy target, according to an AWS Machine Learning Blog post published August 5, 2026. Mobileye, an autonomous driving company with more than 230 million EyeQ system-on-chips deployed across roughly 1,200 vehicle models worldwide, built the agent to reduce routine ticket-status work for skilled engineers.

Mobileye’s Data Collection Processing pipeline ingests thousands of drive-recording sessions daily, creating frequent status inquiries from engineers and data teams. Before the deployment, 66% of support tickets were routine status inquiries, and engineers had to move through 15 clicks across multiple backend systems to identify sessions, check visualization tools, validate outputs, review logs, and write responses.

Mobileye first ran a proof of concept targeting 95% accuracy in ticket classification and sub-2-minute response times. The agent used Anthropic Claude foundation models through Mobileye’s internal LLM Gateway, which provides governed and quota-managed access to foundation models on Amazon Bedrock. Model Context Protocol, or MCP (standardized tool connection for AI agents), let the agent query live drive-data processing APIs during inference, including session status, processing logs, and diagnostic information.

The agent handled more than basic ticket classification. For drive-recording session inquiries, it could confirm completed sessions with access details, surface specific errors with debugging recommendations and log links, or guide users through submission steps for missing requests, all in under two minutes without human intervention.

After the proof of concept, Mobileye chose Amazon Bedrock AgentCore for production because it offered serverless infrastructure, faster support resolution, built-in observability, support for multiple agentic frameworks, and hybrid architecture support. The hybrid design was required because Mobileye’s internal ticketing system ran on-premises and could not be accessed directly from AWS.

The production architecture split work across on-premises systems, AWS services, and Mobileye services. A Local Orchestrator extracted new tickets and posted completed responses back into the internal ticketing system. AgentCore Runtime ran the AI Support Agent with automatic scaling and single-API-call invocation. AgentCore Observability traced each agent interaction, including MCP tool calls, retrieved data, errors, session metrics, latency, token usage, and debugging traces.

AWS Secrets Manager stored, rotated, and retrieved credentials programmatically so teams did not handle secrets directly. Mobileye’s Data Pipeline MCP server gave the agent real-time access to live production data, including driving-session warming statuses, request progress, and processing errors. The AI LLM Gateway managed governed access to foundation models on Amazon Bedrock and other LLM providers.

In production, Mobileye reported a 98% overall success rate, above the original 95% target. Response time fell from hours to about 1 minute, a 90% improvement. The agent automated 66% of total ticket volume and processed 100+ tickets per month.

Mobileye later turned the approach into an internal managed service so developers could deploy production-grade AI agents without AWS expertise or cloud credentials. Teams provide agent code and specify AgentCore capabilities such as Memory, Browser Tool, Code Interpreter, Observability, and Gateway; Mobileye’s Cloud Infra team provisions AWS IAM Roles, Amazon S3 storage, Amazon CloudWatch monitoring, and Amazon Cognito authentication. Developers receive a preconfigured bedrock_agentcore.yaml file and deploy with a single command, agentcore deploy.