Normalized claim
Browser operation tokens: 88% decrease
The team confirmed an 88% reduction in browser operation tokens.
Works Human Intelligence builds Bedrock AgentCore commuting allowance and browser operation agents for HR system automation
Use case typeAI agentsUpdated May 27, 2026
Works Human Intelligence (WHI) develops and supports the integrated HR system COMPANY for major Japanese corporations and public interest corporations. The company built two AI agents with Amazon Bedrock AgentCore to automate routine HR operations, including commuting allowance approvals and browser-based HR system tasks.
- Organization
- Works Human Intelligence
- Industry
- Healthcare
- Location
- Japan
- Published
- May 2026
Reported outcomes
−88%
browser operation tokensOther quantified impact
−97%cost per process
Strategic outcomes
Cost efficiencyReduced monitoring burdenScale & capacityEnabled future sub-agent expansionOther strategic outcomeAutomated HR approval and browser tasks
Why do we believe this?Outcome claims, sources, and evidence checks
Normalized claim
Cost per process: 97% decrease
we succeeded in reducing the cost per process by up to 97%.
Why do we believe this deployment?Customer identity, provider attribution, maturity, and source checks
- Customer
- Works Human Intelligence
- Provider
- AWS
- Maturity
- Production
- Linked source
- AWS Machine Learning Blog
Reduce agent execution cost and operational monitoring overhead
Customer identity supportedSource describes one deploymentMaturity supported
Primary read
Use case focus
Showing 3 of 3
- 1AI agents
- 2Workflow automation
- 3Agent orchestration
- Automate routine HR system operations after employee relocations and other updates.
- Move from a LangGraph proof of concept to an integrated multi-agent architecture.
- Reduce agent execution cost and operational monitoring overhead.
- Built a Commuting Allowance Agent that automates approval workflows and launches sub-agents on separate AgentCore Runtime instances.
- Built a Browser Operation Agent that uses Strands Agents, a knowledge base of operation templates, and browser tools to generate operation manuals, execute HR system changes, and collect evidence.
- Placed AgentCore Runtime inside a VPC with fixed IP access through a NAT gateway, and used Amazon Cognito and AWS Fargate for authentication and authorization.
- Used Amazon Bedrock prompt caching, prompt and behavior optimization, model changes, Amazon S3 for short-term information, and AgentCore Observability for monitoring.
- The Browser Operation Agent achieved an 88% reduction in browser operation tokens.
- Cost per process was reduced by up to 97%.
- Switching from self-hosted Langfuse to AgentCore Observability reduced operational monitoring burden.
- The architecture change enables future expansion of sub-agents and more complex browser workflows.
Architecture
Two AI agents were built on Amazon Bedrock AgentCore. The Commuting Allowance Agent runs sub-agents on separate AgentCore Runtime instances with Amazon Cognito and AWS Fargate for authentication and authorization. The Browser Operation Agent uses Strands Agents, a knowledge base, Amazon S3, and fixed-IP VPC access via a NAT gateway. The workflow also uses Amazon Bedrock prompt caching and AgentCore Observability.
Sources & evidence1
Evidence: Medium50/100Evidence strength
- Customer explicitly identified
- Deployment status explicitly supported
- Quantified outcome available
- Technical implementation details available
Type: Blog PostPublished: May 27, 2026Publisher: AWSEvidence: VendorConfidence: Medium
AI-generated summary. Verify important details with the linked sources before relying on this case.
Explore related AI use cases
Customer evidence
Provider evidence
AI patterns
Browse the catalog
Was this useful?
Community
Comments
No published comments yet.