AWS for Industries
Transforming business operations with multi-agent systems: Field Workforce Safety AI Assistant
Large language models (LLMs) and AI agents have revolutionized the orchestration of complex workflows by using their ability to interact with their environment and reason through actions. Agentic AI solutions bring a new level of flexibility and intelligence, unlike traditional machine learning (ML) methods, which often need complete retraining for new tasks and struggle to adapt to real-time changes. These advanced systems can interact with their environment, reason through actions, and seamlessly connect with external tools, enabling them to handle unpredictable situations and achieve specific goals with remarkable efficiency.
This innovation is particularly impactful in the utilities industry, where the safety of field technicians is a constant concern. Field technicians in the utilities industry carry some of the highest risks for serious injuries or fatalities. Unfortunately, conventional safety management approaches often lack the agility and real-time responsiveness needed to keep up with rapidly changing conditions on the ground.
Multi-agent AI systems are emerging as a game-changing technology that promises to transform business operations across industries. This post explores how multi-agent systems are revolutionizing end-to-end business processes, with a specific focus on their application in utility field operations and worker safety.
What are multi-agent systems?
Multi-agent systems consist of interconnected AI agents that collaborate to solve complex and dynamic problems. Each AI agent operates independently, possessing its own goals, knowledge, and decision-making capabilities. Through communication and coordination, these agents work together to address complex and dynamic problems that need distributed problem-solving and adaptability. This collaborative architecture allows organizations to tackle challenges where cooperation and decentralized decision-making are essential.
AI agents function as independent entities that autonomously assess situations, make decisions, and execute actions to fulfill designated objectives with little to no human intervention. They often have specialized roles, such as monitoring environmental conditions, analyzing data, or enforcing compliance. A coordinating supervisor agent orchestrates the activities that these specialized agents perform, breaking down high-level requests, delegating tasks, and consolidating outputs to make sure of efficient and effective system operation.
This team-based framework implemented by multi-agent systems mirrors the strengths of human collaboration, enabling organizations to benefit from the following:
- Effortless scalability: seamlessly manage increasing workloads and adapt to evolving business needs.
- Automated task handling: offload repetitive processes so that employees can focus on higher-value, strategic activities.
- Enhanced decision-making: integrate and analyze multiple data streams in real-time to support faster, more informed decisions.
- Continuous operation: maintain uninterrupted performance and reliability alongside legacy systems without needing constant human intervention.
In this post, we delve into a shared pursuit within the asset-intensive industry: enhancing the safety of field workforce. We examine how multi-agent systems can transform job safety and risk assessments for field technicians and crews, driving safer outcomes on the ground.
The safety challenge in utilities
Workplace safety remains a persistent challenge across industries, such as utilities, which consistently ranks among the highest-risk sectors for serious injuries and fatalities. In the United States (US), utility workers face a higher risk of fatal accidents when compared to other professions.
Field technicians engaged in infrastructure maintenance and outage response are particularly vulnerable. In 2023, Australia recorded 200 worker fatalities, 5% above the five-year average. In the US, there are approximately 150 electrical work fatalities each year. To reduce risks, regulatory bodies mandate Job Safety Assessments (JSAs) and risk assessments prior to starting work, especially for a field workforce operating in high-risk environments.
Although the specific procedures and documentation differ by country, the fundamental requirement to identify hazards and control risks before work begins is universally upheld and enforced.
- United States: The Occupational Safety and Health Administration (OSHA) necessitates that employers maintain a workplace free from recognized hazards, such as those related to electrical safety.
- European Union: Employers are legally obligated to assess all risks that workers may face, implement preventive and protective measures, and document the entire risk assessment process.
- Australia: Workplace health and safety laws need regular risk assessments and compliance checks, particularly for high-risk activities such as electrical work.
Despite these mandates, field technicians continue to face significant challenges in practice:
- Limited situational awareness: Field technicians often work in unpredictable and hazardous environments, making it difficult to anticipate and respond to dangers.
- Restricted access to real-time data: Without up-to-date information and communication tools, technicians may miss critical safety updates and operational changes, leaving them unprepared for their tasks.
- Balancing customer service with safety compliance: Technicians are expected to deliver prompt service while also adhering to strict safety protocols, which can create conflicting pressures.
- Isolation and delayed emergency response: Working alone in remote areas increases vulnerability and can lead to emergency response delays.
These challenges highlight the need for innovative safety solutions that deliver real-time support and enhance situational awareness. Addressing these challenges allows utilities to better protect field technicians, thus making sure that they can perform their critical roles safely and efficiently.
Solution overview
In this section we will showcase a solution to address field workforce safety challenges. The field workforce safety assistant implements a multi-agent workflow to create a “Job Safety Assessment” based on multiple data sources. This solution uses multiple specialized AI agents, each focused on a specific aspect of safety with all agents working under the coordination of a supervisor agent. We showcase agents that are built using the Amazon Bedrock Multi-Agent collaboration framework and Strand Agents framework.
Through this multi agent collaboration approach, the system delivers comprehensive and context-aware Job Safety Assessments for field teams. The supervisor agent orchestrates the efforts of each specialized agent, making sure that information is integrated and tailored to the unique risks and operational needs of the workforce.
You can view the Field Workforce Safety with multi-agent framework in action by watching the following video.
Figure 1: Field workforce safety with multi-agent framework
Key solution components
The Field Workforce Safety AI Assistant is built on a multi-agent architecture, where each specialized agent is responsible for a distinct aspect of safety. This collaborative approach makes sure that field technicians can receive timely, relevant, and actionable information tailored to their specific tasks and environments. In this workflow, the field technician submits a request for a safety briefing (or a job safety assessment), after which a supervisor agent coordinates with specialized agents to gather and compile the necessary information.
The following figure shows how the Field Workforce Safety AI Assistant uses a supervisor agent to coordinate multiple specialized worker agents. These integrate with external systems to manage and deliver field workforce safety information.
Figure 2. Field Workforce Safety AI Assistant workflow
The following Amazon Web Services (AWS) architecture diagram shows how various AWS services are integrated to deliver the solution.
Figure 3. Field Workforce Safety AI Assistant AWS architecture
The flow depicted in the preceding figure is as follows:
1. Users can also access static content (for example the application) directly from Amazon S3.
2. Users authenticate through Amazon Cognito to access the application.
3. Users request a safety briefing through Amazon API Gateway, which triggers a AWS Lambda function.
4. The Lambda function retrieves the work orders from Amazon DynamoDB (representing your Enterprise Asset Management (EAM) system).
5. The Lambda function invokes the Supervisor Agents (either Amazon Bedrock Agents or Strands Agents) .
6. The Supervisor Agent coordinates with specialized worker agents: Weather Agent, Location Alert Agent, and Site Emergency Management Agent. Each worker agent has been implemented with Lambda.
1. Supervisor agent
The supervisor agent serves as the central coordinator. It analyzes work orders, understands the context, and delegates specific tasks to specialized sub-agents. The multi-agent collaboration capability allows the supervisor to break down complex problems and orchestrate responses.
In supervisor mode, the agent analyzes the input, breaks down complex problems, or rephrases the request. Then, it invokes sub-agents either sequentially or in parallel and may consult knowledge bases or invoke action groups. After receiving responses from sub-agents, the supervisor agent processes them to determine if the problem is solved or if further action is needed.
2. Specialized worker agents
Each worker agent focuses on a specific domain:
- Weather Forecast Agent: fetches weather information at a given location and time to identify environmental risks.
- Safety Officer Agent: analyzes any reported incidents and hazards at the work location to make sure that the technician is well informed.
- Emergency Management Agent: retrieves any emergency notices in the area that could affect operations.
3. Integration components
The system integrates with various data sources to gather the information needed to provide comprehensive safety briefings for field workers:
- Work order details: The system connects to work order management systems to obtain the location and task details for each work order.
- Location-specific hazard information: The system accesses historical incident databases to identify any location-specific hazards or safety issues of which field workers should be aware.
- Real-time environmental alerts: The system integrates with real-time weather and emergency alert services to retrieve any weather, environmental, or other alerts that could impact the work being performed in the field.
Deploy the Field Workforce Safety AI Assistant solution
The field workforce safety AI assistant solution is available as an open source solution at the following GitHub location.
The following steps outline the Field Workforce Safety AI Assistant solution deployment.
1. Prerequisites
- Make sure that you have an AWS account with sufficient permissions to access and manage Amazon Bedrock resources and where you can launch an AWS CloudFormation stack.
- This solution was tested in the us-east-1 (N. Virginia) AWS Region using the Claude 3 Sonnet model on Amazon Bedrock.
- We suggest using this AWS Region and model, but the solution also works in any AWS Region where Amazon Bedrock has access to Claude 3 Sonnet or a more advanced model.
2. Get model access in Amazon Bedrock
A. Sign in to the AWS Management Console and switch to the Region of choice (for example, the us-east-1 Region).
B. In the left navigation pane, under Bedrock configurations, choose Model access.
C. On the Model access page, choose Modify model access.
D. Choose Enable specific models.
E. Scroll to the Anthropic section and check the boxes for Claude 3 Sonnet.
F. Choose Next at the bottom of the page.
G. If prompted, then provide your use case details for Anthropic models and submit the form.
H. Review the terms and conditions, then choose Submit.
I. Wait for access to be granted (the status updates to Access granted).
For more information view the Access Amazon Bedrock foundation models user guide.
3. Find model IDs
A. In the Amazon Bedrock console, under Foundation models, choose Model catalog from the sidebar.
B. Locate and choose Claude 3 Sonnet.
Figure 4. Amazon Bedrock Model Catalog
C. In the details section, copy the Model ID (you use this later).
Figure 5. Amazon Bedrock Model Details
4. Deploy the CloudFormation stack
A. Select Launch Stack.
B. On the Specify stack details page:
a. Enter a stack name (for example FieldWorkForceSafetyMainStack).
b. For CollaboratorFoundationModel, leave the default value for Claude 3 Sonnet model ID.
c. For SupervisorFoundationModel, leave the default value for Claude 3 Sonnet model ID.
C. Choose Next after entering the stack details.
D. On Configure stack options, scroll down, review the AWS Identity and Access Management (IAM) role information, and check I acknowledge that AWS CloudFormation might create IAM resources.
Figure 6. Amazon CloudFormation Stack Creation
E. On Review and create, scroll down and choose Submit.
F. Stack creation begins and may take 20–30 minutes to complete.
5. Access the solution
A. After deployment, choose the created stack FieldWorkForceSafetyMainStack in CloudFormation.
B. In the Outputs tab, find and choose the FrontendUrl to open the app (for example https://d111111abcdef8.cloudfront.net).
Figure 7. Amazon CloudFormation Stack Outputs
C. Choose Create account.
D. Enter your email address as the username and set a password.
Figure 8. Field Workforce Safety Assistant Create Account Page
E. Check your email for a six-digit verification code, enter it, and choose Confirm.
F. The app supports allows you to switch between Strands agent framework (default) or Amazon Bedrock Agents. To view the current framework in use, click on the logged in username and select Profiles and Settings.
Figure 9. Field Workforce Safety Assistant Profiles & Settings
G. On the User Profile & Settings dialog box, notice the Strands AI agents is set as default, however you can switch to Amazon Bedrock Agents. Leave the default as Strands AI Agents, Save preferences and Close dialog box.
Figure 10. Alter the AI Agent framework
H. On the Work Order queue screen, choose a Work order.
Figure 11. Field Workforce Safety Assistant Work Order Page
I. Choose Load emergency warnings to fetch new warnings around the site.
Figure 12. Field Workforce Safety Assistant Work Order Emergency Warnings Map
J. Choose Perform safety check and scroll down to view the Job Safety Assessment.
Figure 13. Field Workforce Safety Assistant Work Order Safety Briefing
Future opportunities
As multi-agent systems continue to evolve, their impact extends far beyond field safety, offering transformative possibilities for automating complex, end-to-end business processes across electrical utilities. For example, responding to a network maintenance or outage management event is a multifaceted challenge that involves several distinct business functions, each with its own responsibilities and expertise. Multi-agent collaboration is uniquely suited to this complexity—deploying specialized agents for each function allows utilities to automate and coordinate every step of the workflow.
For example, consider an outage management scenario involving five key areas:
1. Grid operations and control: This group of agents is responsible for monitoring grid stability, forecasting weather-related disruptions, and coordinating responses across operational teams.
2. Maintenance planning and scheduling: Agents in this area optimize resource allocation, manage inventory, and develop maintenance schedules to maximize efficiency and minimize service disruptions.
3. Field crew operations: These agents deliver real-time safety updates, provide technical guidance, and automate compliance documentation to support field crews during operations.
4. Customer communications: Agents in this section manage proactive customer notifications, analyze feedback to enhance service quality, and customer issues are properly addressed and closed.
5. Analytics and planning: This group evaluates maintenance effectiveness, predicts equipment failures, and recommends infrastructure investments to prevent future problems.
In this workflow, each business unit manages its specialized agents, while the overall process is orchestrated through seamless, automated handoffs with humans in-the-loop. This breaks down silos and creates an integrated system that’s more efficient than the sum of its parts. Supervising agents can autonomously execute multi-step processes across multiple business units, streamlining operations, reducing manual effort, and improving efficiency, quality, and customer satisfaction. As businesses add new agents, they can be reused across different systems, further streamlining business processes with minimal added development effort.
Moreover, this vision isn’t just theoretical—organizations are already implementing multi-agent systems to manage complex workflows across previously siloed operations. Instead of building monolithic applications that try to do everything, businesses are creating specialized agents that excel at specific tasks while collaborating seamlessly through standardized protocols.
Conclusion
Multi-agent systems will revolutionize business operations, particularly in high-risk environments such as utility field operations, where safety and efficiency are critical. Aligning specialized AI agents with functional areas within end-to-end business processes allows organizations to create intelligent workflows that enhance collaboration and create intelligent, adaptive workflows.
The Field Workforce Safety AI Assistant demonstrates how the multi-agent collaboration (implemented using Amazon Bedrock agents or Strands agents) capabilities can be applied to address real-world safety challenges and protect front-line electrical workers. As these technologies continue to evolve, we can anticipate improvements that further improve safety, operational efficiency, and customer satisfaction.
We’d love to hear from you. Let us know what you think in the comments section, or use the issues forum in the Field Workforce safety AI assistant GitHub repository.
To learn more about building multi-agent systems visit the Amazon Bedrock documentation (sample implementations) or Strands Agents (sample implementation).