How Swisscom builds enterprise agentic AI for buyer assist and gross sales utilizing Amazon Bedrock AgentCore

This put up was written with Arun Sittampalam and Maxime Darcot from Swisscom.

As we navigate the consistently shifting AI ecosystem, enterprises face challenges in translating AI’s potential into scalable, production-ready options. Swisscom, Switzerland’s main telecommunications supplier with an estimated $19B income (2025) and over $37B Market capitalization as of June 2025 exemplifies how organizations can efficiently navigate this complexity whereas sustaining their dedication to sustainability and excellence.

Swisscom has been acknowledged because the Most Sustainable Firm within the Telecom business for 3 consecutive years by World Finance journal, Swisscom has established itself as an innovation chief dedicated to reaching net-zero greenhouse gasoline emissions by 2035 in alignment with the Paris Local weather Settlement. This sustainability-first strategy extends to their AI technique the place they’re breaking by way of what they name the “automation ceiling” – the place conventional automation approaches fail to fulfill trendy enterprise calls for.

On this put up, we’ll present how Swisscom carried out Amazon Bedrock AgentCore to construct and scale their enterprise AI brokers for buyer assist and gross sales operations. As an early adopter of Amazon Bedrock within the AWS Europe Area (Zurich), Swisscom leads in enterprise AI implementation with their Chatbot Builder system and varied AI initiatives. Their profitable deployments embody Conversational AI powered by Rasa and fine-tuned LLMs on Amazon SageMaker, and the Swisscom Swisscom myAI assistant, constructed to fulfill Swiss information safety requirements.

Resolution overview: Swisscom’s agentic AI enabler framework

The problem of enterprise-wide scaling of AI brokers lies in managing siloed agentic options whereas facilitating cross-departmental coordination. Swisscom addresses this by way of Mannequin Context Protocol (MCP) servers and the Agent2Agent protocol (A2A), for seamless agent communication throughout domains. Working beneath Switzerland’s strict information safety legal guidelines, they’ve developed a framework that balances compliance necessities with environment friendly scaling capabilities, serving to stop redundant efforts whereas sustaining excessive safety requirements.

Swisscom’s multi-agent structure: System design and implementation challenges

Swisscom’s imaginative and prescient for enterprise-level agentic AI focuses on addressing basic challenges that organizations face when scaling AI options. They recognise that profitable implementation requires extra than simply revolutionary know-how, it calls for a complete strategy to infrastructure and operations. One of many key challenges lies in orchestrating AI brokers throughout totally different departments and techniques whereas sustaining safety and effectivity.

For example these challenges in apply, let’s look at a typical customer support state of affairs the place an agent is tasked with serving to a buyer restore their Web router connectivity. There are three potential causes for the connectivity loss: 1) a billing situation, 2) a community outage, or 3) a configuration mismatch referred to as a pairing situation. These points usually reside in departments totally different from the place the assigned agent operates, highlighting the necessity for seamless cross-departmental coordination.

The structure diagram under illustrates the imaginative and prescient and related challenges for a generic buyer agent with out the Amazon Bedrock AgentCore. The shared VPC setup of Swisscom is defined in additional element within the weblog put up, Automated networking with shared VPCs at Swisscom.

This structure consists of the next parts:

1. A customer-facing generic agent deployed as a containerized runtime inside a shared VPC, requiring each basis mannequin invocation capabilities and strong session administration.
2. For job completion, the agent requires to entry to different brokers and MCP servers. These assets are usually distributed throughout a number of AWS accounts and are deployed as containerised runtimes throughout the shared VPC.
3. Inside software entry primarily happens by way of SAIL (Service and Interface Library), Swisscom’s central system for API internet hosting and repair integration. Company community assets are accessible through AWS Direct Join, with a VPC Transit Gateway facilitating safe cross-network communication.
4. Safety compliance is paramount: every interplay requires non permanent entry tokens that authenticate each the agent and the shopper context. This bidirectional validation is crucial to the system parts – brokers, MCP servers, and instruments should confirm incoming tokens for service requests.
5. Gaining long-term insights from the saved classes, akin to buyer preferences, calls for a complicated evaluation.

To construct the answer talked about above at scale, Swisscom recognized a number of crucial challenges that wanted to be addressed:

• Safety and Authentication:
  • The best way to implement safe, transitive authentication and authorization that enforces least-privilege entry primarily based on intersecting permissions (buyer, agent, division)?
  • The best way to allow managed useful resource sharing throughout departments, cloud techniques, and on-premises networks?
• Integration and Interoperability:
  • The best way to make MCP servers and different brokers centrally obtainable to different use instances?
  • The best way to combine and preserve compatibility with present agentic use instances throughout Swisscom’s infrastructure?
• Buyer Intelligence and Operations:
  • The best way to successfully seize and make the most of buyer insights throughout a number of agentic interactions?
  • The best way to implement standardized analysis and observability practices throughout the brokers?

How Amazon Bedrock AgentCore addresses the challenges

Amazon Bedrock AgentCore supplies Swisscom with a complete resolution that addresses their enterprise-scale agentic AI challenges.

• AgentCore Runtime: Permits Swisscom’s builders to concentrate on constructing brokers whereas the system handles safe, cost-efficient internet hosting and computerized scaling by way of Docker container deployment that maintains session-level isolation. Hosted within the shared VPC permits entry to inside APIs.
• AgentCore Id: Seamlessly integrates with Swisscom’s present identification supplier, managing each inbound and outbound authentication, assuaging the necessity for customized token alternate servers and simplifying safe interactions between brokers, instruments, and information sources.
• AgentCore Reminiscence: Delivers a sturdy resolution for managing each session-based and long-term reminiscence storage with customized reminiscence methods. That is significantly invaluable for B2C operations the place understanding buyer context throughout interactions is essential. Conserving every consumer’s information separate additionally helps safety and compliance efforts.
• Strands Brokers Framework: Demonstrates excessive adoption amongst Swisscom’s builders attributable to its simplified agent building, quicker improvement cycles, seamless integration with Bedrock AgentCore companies, and built-in capabilities for tracing, analysis, and OpenTelemetry logging.

This resolution does the next:

1. The shopper sends a request to the Strands agent working on AgentCore Runtime, passing an authentication token from the Swisscom IdP.
2. The shopper’s token is validated and a brand new token for the agent’s downstream instrument utilization is generated and handed again to the agent.
3. The agent invokes the muse mannequin on Bedrock and shops the classes within the AgentCore Reminiscence. The visitors traverses the VPC endpoints for Bedrock and Bedrock AgentCore, protecting the visitors non-public.
4. The agent accesses inside APIs, MCP & A2A servers contained in the shared VPC, authenticating with the non permanent token from AgentCore Id.

With the flexibleness to make use of a subset of options of Amazon Bedrock AgentCore and their Amazon VPC integration Swisscom may stay safe and versatile to make use of the Bedrock AgentCore companies for his or her particular wants, for instance to combine with present brokers on Amazon EKS. Amazon Bedrock AgentCore integrates with VPC to facilitate safe communication between brokers and inside assets.

Outcomes and advantages: Actual-world implementation with self-service use case

Swisscom partnered with AWS to implement Amazon Bedrock AgentCore for 2 B2C instances: 1) producing customized gross sales pitches, and a couple of) offering automated buyer assist for technical points like self-service troubleshooting. Each brokers are being built-in into Swisscom’s present buyer generative AI-powered chatbot system known as SAM, necessitating high-performance agent-to-agent communication protocols because of the excessive quantity of Swisscom clients and strict latency necessities. All through the event course of, the staff created an agent for every use case designed to be shared throughout the group by way of MCP and A2A.

Amazon Bedrock AgentCore has confirmed instrumental in these implementations. By utilizing the Bedrock AgentCore Reminiscence long-term insights Swisscom can observe and analyze buyer interactions throughout totally different touchpoints, constantly bettering the shopper expertise throughout domains. AgentCore Id facilitates strong safety, implementing exact entry controls that restrict brokers to solely these assets licensed for the particular buyer interplay. The scalability of AgentCore Runtime permits these brokers to effectively deal with 1000’s of requests per thirty days every, sustaining low latency whereas optimizing prices.

The adoption of Strands Brokers framework has been significantly invaluable on this journey:

• Growth groups achieved their first enterprise stakeholder demos inside 3-4 weeks, regardless of having no prior expertise with Strands Brokers.
• One challenge staff migrated from their LangGraph implementation to Strands Brokers, citing decreased complexity and quicker improvement cycles.
• The framework’s native OpenTelemetry integration supported seamless export of efficiency traces to Swisscom’s present observability infrastructure, sustaining consistency with enterprise-wide monitoring requirements.
• The Strands analysis take a look at instances allowed groups rapidly put an analysis pipeline collectively with out the necessity of further instruments, for a fast validation of the PoC.

Conclusion: Enterprise AI at scale – Key insights and Strategic implications

Swisscom’s implementation of Amazon Bedrock AgentCore demonstrates how enterprises can efficiently navigate the complexities of production-ready Agentic AI whereas sustaining regulatory compliance and operational excellence. Swisscom’s journey provides 3 crucial insights:

• Architectural basis issues: By addressing the basic challenges of safe cross-org authentication, standardized agent orchestration, and complete observability, Swisscom established a scalable basis that accelerates deployment somewhat than constraining it. The mixing of AgentCore Runtime, Id, and Reminiscence companies accelerated the infrastructure setup so groups may concentrate on enterprise worth.
• Framework choice drives velocity: The adoption of Strands Brokers framework exemplifies how the appropriate improvement instruments can dramatically scale back time-to-value. Groups reaching stakeholder demos inside 3-4 weeks, coupled with profitable migrations from different frameworks, validates the significance of developer expertise in enterprise AI adoption.
• Compliance as an enabler: Swisscom proved that regulatory compliance needn’t impede innovation. The system’s capability to scale whereas sustaining information sovereignty and consumer privateness has confirmed significantly invaluable within the Swiss business, the place regulatory compliance is paramount.

As enterprises more and more acknowledge AI brokers as basic to aggressive benefit, Swisscom’s implementation supplies a confirmed reference structure. Their success with high-volume B2C purposes—from customized gross sales help to automated technical assist—illustrates that agentic AI can ship measurable enterprise outcomes at scale when constructed on acceptable infrastructure. This implementation serves as a blueprint for organizations looking for to deploy enterprise-scale AI options, exhibiting how cautious architectural planning and the appropriate know-how selections can result in profitable outcomes in each customer support and gross sales operations.

Subsequent steps and searching forward

The long run roadmap focuses on three key areas: agent sharing, cross-domain integration, and governance. A centralized agent registry will facilitate discovery and reuse throughout the group, supported by standardized documentation and shared greatest practices. Cross-domain integration will allow seamless collaboration between totally different enterprise models, with clear requirements for agent communication and interoperability. The implementation of sturdy governance mechanisms, together with model management, utilization monitoring, and common safety audits, will facilitate sustainable development of the system whereas sustaining compliance with enterprise requirements. This complete strategy will assist drive steady enchancment primarily based on real-world utilization patterns and suggestions.

Try these further hyperlinks for related Agentic associated info:

In regards to the authors

Arun Sittampalam, Director of Product Administration AI at Swisscom, leads the corporate’s transformation towards Agentic AI, designing frameworks that scale massive language mannequin (LLM)–pushed brokers throughout enterprise environments. His staff is constructing Swisscom’s agentic platform, integrating Amazon Bedrock, AgentCore and inside orchestration frameworks to empower Swisscom’s AI product groups to construct and scale clever brokers quicker. Arun focuses on operationalizing multi-agent architectures that ship automation, reliability, and scalability.

Maxime is a System and Safety Architect at Swisscom, answerable for the structure of Conversational and Agentic AI enablement. He’s initially a Information Scientist with 10 years of expertise in creating, deploying and sustaining NLP options which have been serving to hundreds of thousands of Swisscom clients.

Julian Grüber is a Information Science Advisor at Amazon Internet Providers. He companions with strategic clients to scale GenAI options that unlock enterprise worth, working at each the use case and enterprise structure stage. Drawing on his background in utilized arithmetic, machine studying, enterprise, and cloud infrastructure, Julian bridges technical depth with enterprise outcomes to handle complicated AI/ML challenges.

Marco Fischer is a Senior Options Architect at Amazon Internet Providers. He works with main telecom operators to design and deploy scalable, production-ready options. With over 20 years of expertise spanning software program engineering, structure, and cloud infrastructure, Marco combines deep technical experience with a ardour for fixing complicated enterprise challenges.

Akarsha Sehwag is a Generative AI Information Scientist for Amazon Bedrock AgentCore GTM staff. With over six years of experience in AI/ML, she has constructed production-ready enterprise options throughout numerous buyer segments in Generative AI, Deep Studying and Laptop Imaginative and prescient domains. Exterior of labor, she likes to hike, bike or play Badminton.

Ruben Merz is a Principal Options Architect at AWS, specializing in digital sovereignty, AI, and networking options for enterprise clients. With deep experience in distributed techniques and networking, he architects safe, compliant cloud options that assist organizations navigate complicated regulatory necessities whereas accelerating their digital transformation journeys.