Building Agentic Workflows with IBM watsonx Orchestrate on AWS

Enterprises deploying AI agents across business functions are working to unify automation, strengthen governance, and simplify integration. To deliver business value, AI agents require the ability to coordinate across systems and maintain consistent security controls. Achieving this requires orchestration capabilities that enable agents to work together effectively, reducing manual processes and extending AI investments across workflows.

From financial services and manufacturing to retail and healthcare, these enterprises are implementing agentic workflows spanning multiple business functions—from procurement and HR to sales, customer service, and other operational areas. These workflows must connect AI capabilities with existing business systems while maintaining governance and auditability.

Available through AWS Marketplace, IBM watsonx Orchestrate provides AI orchestration capabilities that combine structured automation with AI-driven decision-making. The solution integrates with Amazon Bedrock foundation models and Amazon Bedrock AgentCore services to help organizations automate business processes while maintaining security features.

In this post, you will learn implementation options for building agentic workflows with watsonx Orchestrate on AWS. You will explore visual builders for business users and development toolkits for developers, supporting collaboration across your organization.

IBM watsonx Orchestrate on AWS

IBM watsonx Orchestrate is designed to help organizations build, deploy, and manage AI agents across business functions. The solution provides natural language interfaces through chat and voice interactions.

Core capabilities

Orchestration engine: Routes requests to prebuilt or custom agents and coordinates workflow execution across systems. The engine connects to data sources and manages task sequences across your business applications.

Agent and tool catalog: Includes prebuilt agents for HR, procurement, finance, sales, and customer service. The catalog provides domain-specific agents and tools that connect to enterprise systems. You can deploy these agents directly or customize them for specific requirements.

Development interfaces: Offers two paths for building automation. Business users access visual builders with drag-and-drop interfaces to create workflows without coding. Developers use the IBM watsonx Orchestrate Agent Development Kit (ADK) to build custom agents using Python, OpenAPI specifications, and Langflow.

AI Gateway: Connects to foundation models from Amazon Bedrock and other providers. You select models based on task requirements such as natural language understanding, reasoning, or content generation.

Enterprise integration layer: Links to business systems through connections with enhanced security features. The solution includes prebuilt integrations for Salesforce, SAP, Workday, and AWS services.

Agentic workflows with watsonx Orchestrate

Agentic workflows combine agents, tools, and people to fulfill user requests using both deterministic and AI-driven automation. These workflows define how work gets done for business processes such as procure-to-pay or lead-to-order, ensuring agents execute consistently with accuracy, auditability, and compliance.

Workflows in watsonx Orchestrate include:

Predefined toolchains: Specify interaction patterns between agents and enterprise systems with defined sequences of tool invocations.
Execution logic: Include conditionals and branching paths that determine workflow progression based on data conditions and business rules.
Data handling: Transform information between workflow steps. Map inputs and outputs across different systems and formats.
Context-aware agent capabilities: AI agents make decisions based on conversation context, user intent, and available tools.

The following diagram shows a watsonx Orchestrate workflow that sequences tool calls, AI model invocations, and human interactions with conditional branching and automatic data mapping to calculate insurance rates based on either VIN numbers or vehicle make and model (Figure 1):

Screenshot of the IBM watsonx Orchestrate agentic workflow builder with a workflow that sequences tool calls, AI model invocations, and human interactions with conditional branching and automatic data mapping to calculate insurance rates based on either VIN numbers or vehicle make and model.

Figure 1. IBM watsonx Orchestrate agentic workflow with tool calls, AI models, and human interactions.

Business value

Organizations use agentic workflows to:

Automate dynamic processes: Handle complex scenarios with loops for iterative tasks, conditional branching for decision logic, and document processing for extracting information from unstructured data.
Integrate with business systems: Connect to existing applications through APIs. Agents retrieve and update records across different business applications like ERP, CRM.
Support agent collaboration: Configure agent-to-agent collaboration where specialized agents handle specific tasks. Import external tools through standardized protocols including Model Context Protocol (MCP) and OpenAPI.

These capabilities help you reduce manual effort in repetitive tasks, improve consistency across business processes, and maintain audit trails for compliance requirements. You can add human checkpoints at workflow stages where judgment is required, such as high-value approvals or compliance reviews.

Development paths for building agentic workflows

IBM watsonx Orchestrate provides two development paths that support collaboration between technical and business teams.

For developers: Agent Development Kit

ADK provides code-based development using Python, OpenAPI specifications, and support for multiple frameworks such as Langflow. Developers build custom agents and tools, create data transformations, and integrate with APIs using their preferred development environment.

It integrates with Git for version control and CI/CD pipelines for automated deployment. You define agent logic, configure tool access, and deploy agents to watsonx Orchestrate through the ADK command-line interface. Custom agents built with the ADK appear in the watsonx Orchestrate catalog, making them available to business users through the visual builder.

For business users: Visual workflow builder

The visual builder provides drag-and-drop interfaces with prebuilt components for creating workflows without coding requirements. You design workflows using guided steps and templates, configure conditional logic and data mapping, and set up approval routing.

The visual builder includes testing capabilities to validate that workflows meet technical and operational requirements before deployment. You modify deployed workflows through the same interface, with changes tracked for audit purposes. Business users can incorporate custom agents and tools built by developers into their workflows through the catalog.

Use case for prebuilt agents: Procurement workflow automation

The following example demonstrates how prebuilt procurement agents from the watsonx Orchestrate catalog automate invoice processing workflows. This workflow coordinates activities across multiple systems and stakeholders, combining AI-driven validation with deterministic business rules and human oversight.

An invoice processing workflow includes the following steps:

Document extraction: Document processing tools extract vendor details, invoice numbers, line items, amounts, and payment terms from invoices received through email or electronic submission channels.
Validation: The agent queries purchase order systems through API connections to validate invoice information against existing records. The agent compares invoice line items with purchase order details and receiving reports to identify discrepancies.
Policy application: Business rules evaluate the invoice against policies based on amount thresholds, vendor categories, and contract terms. The workflow applies different approval routes based on invoice value and vendor status.
Approval routing: The workflow routes invoices to appropriate approvers based on configured business rules. High-value invoices trigger human approval checkpoints where finance managers review and approve before payment processing.
Payment processing: For approved invoices, the system generates payment instructions and updates accounting systems through API connections to ERP platforms. The workflow creates payment records and schedules payment dates based on invoice terms.
Audit trail maintenance: The workflow logs all actions, decisions, and approvals with timestamps and user identifiers. This audit trail supports compliance reporting and provides visibility into payment processing timelines.

This workflow shows how watsonx Orchestrate combines AI-driven validation with deterministic business rules, human oversight, and system integration to automate end-to-end procurement processes while maintaining accuracy and compliance.

Integration with AWS services

IBM watsonx Orchestrate integrates with AWS services to extend agentic workflow capabilities. While the previous section focused on a specific procurement workflow, the following architecture diagram shows the broader integration options between watsonx Orchestrate, Amazon Bedrock AgentCore, and Amazon Bedrock (Figure 2):

Image of a reference architecture diagram showing the integrations between AI Agents on IBM watsonx Orchestrate SaaS on AWS, with AWS services like Amazon Bedrock AgentCore using MCP and A2A protocols. And integrations with Amazon Bedrock for foundation models and AI Agents.

Figure 2. IBM watsonx Orchestrate integration points with AWS services.

Architecture walkthrough

The following walkthrough explains the components from the architecture diagram (Figure 2) and how they work together to build, deploy, and monitor AI agents:

AgentCore Runtime provides serverless hosting for agents using open-source AI frameworks like Strands, LangGraph, CrewAI and others. It supports Model Context Protocol (MCP) for standardized tool access and Agent-to-Agent (A2A) for multi-agent coordination. User sessions run in dedicated, isolated environments that persists for up to 8 hours, enabling multi-step workflows while maintaining security. For agent reasoning and decision-making, you can use any foundation models from Amazon Bedrock, OpenAI, or other model providers.
AgentCore Gateway transforms AWS Lambda functions into MCP-compatible tools. AWS Lambda functions contain business logic and integrate with AWS services like Amazon DynamoDB. AgentCore Gateway handles protocol translation and uses AWS Identity and Access Management (IAM) execution roles to invoke AWS Lambda functions.
AgentCore Identity manages inbound authentication from watsonx Orchestrate, integrating with Amazon Cognito to validate OAuth tokens for AgentCore Runtime and Gateway access.
AgentCore Memory enables agents to store and retrieve both short-term session context and long-term user insights, allowing them to deliver personalized, context-aware conversations and build understanding of users over time.
AgentCore Observability provides real-time visibility into agent performance through Amazon CloudWatch dashboards and telemetry, enabling developers to trace execution paths, debug bottlenecks, and maintain quality standards in production.
Amazon Bedrock provides foundation models such as Anthropic Claude or Amazon Nova for agents deployed to AgentCore Runtime or watsonx Orchestrate. Generate Amazon Bedrock API keys and IAM access keys to enable model access in watsonx Orchestrate.
Amazon Bedrock Agents use retrieval-augmented generation (RAG) with Amazon Bedrock Knowledge Bases and foundation models to answer domain-specific questions.
AWS Lambda functions expose Amazon Bedrock Agents through Amazon API Gateway, creating REST API endpoints compatible with the OpenAI chat completions format. This allows watsonx Orchestrate to invoke Bedrock Agents directly as collaborators without requiring custom code.
Create connections in watsonx Orchestrate, for both draft and live environments to securely access AWS services. Configure three connection types: OAuth tokens for Amazon Bedrock AgentCore Runtime and Gateway (inbound authentication), API keys for Amazon API Gateway endpoints, and API keys with IAM credentials for Amazon Bedrock foundation models.
Add Amazon Bedrock foundation models to the watsonx Orchestrate AI Gateway, making them available in Agent Builder and ADK for agent development. The following screenshot shows Amazon Bedrock foundation models in Agent Builder (Figure 3):

Screenshot of IBM watsonx Orchestrate showing Amazon Bedrock foundation models available in watsonx Orchestrate Agent Builder.

Figure 3. Amazon Bedrock foundation models available in watsonx Orchestrate Agent Builder.

Discover and import MCP tools from AgentCore Gateway and AgentCore Runtime. Agents from Amazon Bedrock AgentCore Runtime and Amazon Bedrock can be used as collaborators via A2A protocol or OpenAI chat completions endpoint (Figure 4):

Screenshot of IBM watsonx Orchestrate Agent Builder tool showing tools imported from AgentCore Runtime MCP server available in watsonx Orchestrate.

Figure 4. Imported tools from AgentCore Runtime MCP server available in watsonx Orchestrate Agent Builder.

After building your agents, deploy them to channels such as the watsonx Orchestrate chat interface (Figure 5). A supervisor agent receives user requests and routes them to specialized agents based on task context. These agents coordinate with AWS services—including Amazon Bedrock AgentCore services, AWS Lambda functions, and Amazon Bedrock models—or operate independently to deliver responses.

Screenshot of IBM watsonx Orchestrate shows how users can interact with AI agents in the chat interface to automate tasks, manage workflows, and coordinate actions across multiple agents.

Figure 5. Use AI agents in the chat interface to automate tasks, manage workflows, and coordinate actions across different systems.

This architecture walkthrough explained how watsonx Orchestrate agents can be integrated with Amazon Bedrock AgentCore managed services for secure, scalable agent operations. For implementation details, see the Get started with Amazon Bedrock AgentCore and IBM watsonx Orchestrate documentation.

Summary

In this post, you learned how IBM watsonx Orchestrate helps organizations build and deploy agentic workflows on AWS. You explored implementation options including the visual workflow builder for business users and the Agent Development Kit for developers. You also learned how watsonx Orchestrate integrates with Amazon Bedrock foundation models, Amazon Bedrock AgentCore services, and Amazon Bedrock Agents to coordinate AI agents across business functions.

To get started with watsonx Orchestrate, visit the AWS Marketplace listing below or explore additional IBM solutions on AWS through the resources provided.

AWS Marketplace:

Additional Content:

IBM & Red Hat on AWS