Streamlining HR Tasks: Developing AI Agents with Works Human Intelligence and AWS
Introduction to AI in HR
Developing AI agents for business support presents unique challenges that many organizations face when trying to automate routine HR tasks. Works Human Intelligence (WHI) develops, sells, and supports the integrated HR system “COMPANY” for major Japanese corporations and public interest corporations.
Collaboration with AWS Generative AI Innovation Center
In this post, we share how the AWS Generative AI Innovation Center (GenAIIC) collaborated with Works Human Intelligence (WHI) to build two AI agents using Amazon Bedrock AgentCore. We discuss the challenges encountered and the solutions that reduced costs by up to 97% while improving operational efficiency.
Addressing Common HR Challenges
Customers using HR systems must respond to numerous situations, such as organizational changes, revisions to HR systems, and updates to employee information. AI agents can significantly reduce workload and improve productivity by addressing similar challenges.
Project Overview
The project encompasses two AI agents designed to support operational departments: the Commuting Allowance Agent and the Browser Operation Agent. This post will discuss the challenges and solutions for both agents.
1. Commuting Allowance Agent
- Automating Approvals: This agent automates the approval of applications during employee relocations.
- Challenges: Transitioning from a monolithic configuration to a multi-agent environment, integrating with AWS services for authentication and authorization.
- Solutions and Architecture: Revised architecture for sub-agent deployment, enhancing tenant management with Amazon DynamoDB and Amazon Cognito.
2. Browser Operation Agent
- Enhancing Operations: This agent interacts with the HR system, performing operations and gathering necessary information.
- Challenges: Reducing tokens in browser operations and migrating to more flexible agent structures.
- Solutions and Architecture: Built using Strands Agents, with an optimized workflow and enhanced IP access configuration involving a VPC.
Results & Impact
Both agents demonstrated significant efficiency improvements, reducing costs dramatically while handling complex tasks more effectively.
Conclusion
Our collaboration has positioned WHI to focus on developing business logic while benefiting from the streamlined processes offered by Amazon Bedrock’s AgentCore. We are excited about the future potential for AI agents in HR.
About the Authors
Meet the team behind this innovative project, composed of experienced leaders in generative AI and cloud solutions. Each member brings unique insights and expertise to this transformative venture in HR automation.
Developing AI Agents for Business Support: A Collaboration Between AWS and Works Human Intelligence
In the world of business, efficiency is everything. As organizations seek to streamline operations, many are turning to AI to automate routine Human Resources (HR) tasks. However, developing AI agents that genuinely support these initiatives can be complex. Works Human Intelligence (WHI) has taken significant strides in this arena by creating an integrated HR system called “COMPANY” for major Japanese corporations and public interest entities.
In this blog post, we’ll explore how the AWS Generative AI Innovation Center (GenAIIC) collaborated with WHI to build two innovative AI agents using Amazon Bedrock AgentCore. We’ll discuss the challenges faced during development and highlight how our solutions not only improved operational efficiency but also reduced costs by up to 97%.
The Need for AI Agents in HR
Organizations constantly face dynamic situations including organizational changes, updates to HR systems, and modifications in employee information. The implementation of AI agents can significantly alleviate the workload for HR professionals, leading to enhanced productivity. WHI recognized this potential and set out to develop products utilizing AI agents. However, challenges soon emerged.
Key Challenges
As WHI embarked on building its AI solutions, several hurdles presented themselves:
- The existing proof of concept (PoC) for the Commuting Allowance Agent was built using LangGraph and Amazon ECS, but faced limitations due to monolithic configurations.
- For the Browser Operation Agent, reliance on proprietary implementations created migration difficulties and inefficiencies in operations.
In response to these issues, the GenAIIC team collaborated closely with WHI to foster innovative solutions.
Commuting Allowance Agent
The Commuting Allowance Agent automates the approval of commuting allowance applications, especially during employee relocations.
Challenge
WHI was already developing the Commuting Allowance Agent using LangGraph. However, the release of Amazon Bedrock AgentCore led them to reconsider their architecture. They needed a flexible multi-agent environment with robust authentication and authorization features.
Solution Overview
We shifted the architecture to launch sub-agents independently on the AgentCore Runtime. With multi-tenancy support integrated through Amazon DynamoDB and Amazon Cognito, WHI gained greater flexibility.
Architecture
The Commuting Allowance Agent interfaces through Slack, allowing seamless authentication and operation. The new architecture enabled efficient sub-agent processing, paving the way for future enhancements.
Results and Impact
Utilizing Amazon Bedrock AgentCore improved development times significantly. With this new setup, WHI reduced operational burdens and streamlined their development processes, making it easier to create new sub-agents in the future.
Browser Operation Agent
This agent accesses the “COMPANY” HR system to perform operations and collect necessary information on behalf of users.
Challenge
Initially developed with LangGraph and Playwright Model Context Protocol (MCP), the team faced issues related to token consumption and migrant integration complexities with Strands Agents.
Solution Overview
We built the Browser Operation Agent using Strands Agents and developed a streamlined workflow to minimize token usage. This included searching for optimal operation templates based on user instructions.
Architecture
To manage restricted access from the agent to “COMPANY,” we placed the AgentCore Runtime within a virtual private cloud (VPC) and configured access through a NAT gateway, ensuring secure operations and data flow.
Results and Impact
By optimizing token usage and leveraging prompt caching, the team achieved an impressive reduction in processing costs. Collaborative improvements brought about significant operational efficiencies, including the ability to execute multiple changes seamlessly.
Conclusion
Our collaboration with WHI successfully transitioned the AI agent execution to the AgentCore Runtime, allowing for real-time operational checking through AgentCore Observability. WHI has reported that this transition has fueled development agility and minimized implementation complexities.
Through building these AI agents, WHI can now focus on enhancing business logic while enjoying the foundational support provided by AgentCore.
If you’re curious about how Amazon Bedrock AgentCore can simplify AI agent development for your organization, check out our Getting Started Guide or explore the Hands-on Lab. With the right tools, you can automate routine tasks, build multi-agent workflows, and optimize costs effectively.
About the Authors
Dayuan Jiang is a Senior Deep Learning Architect at AWS in Tokyo, specializing in generative AI technologies.
Minsup Sim serves as a Deep Learning Architect at the AWS Generative AI Innovation Center, focusing on end-to-end generative AI solutions.
Angie Wang is a Senior Generative AI Strategist dedicated to accelerating generative AI adoption across the Asia-Pacific region.
Toshio Katsurai is a Senior Solutions Architect in the ISV/SaaS team at AWS Japan, helping establish secure and optimized environments on AWS.
Together, we’re committed to driving innovation and efficiency in the realm of AI-driven business solutions.