Building the Bot Factory: Scaling AI Agents at AutoScout24
From Disparate Experiments to a Standardized Framework
The Challenge: Identifying a High-Impact Use Case
Architectural Overview
The Solution: A Reusable Blueprint Powered by AWS
Orchestrator Agent: Built with Strands SDK
The Impact: A Validated Blueprint for Enterprise AI
Conclusion: A New Model for Enterprise Agents
About the Authors
AutoScout24: Pioneering AI with a Bot Factory
AutoScout24 has established itself as Europe’s leading automotive marketplace platform, connecting buyers and sellers of new and used vehicles across multiple countries. But as the marketplace grows, so does the complexity of its operations. This realization has led AutoScout24 to embark on a transformative journey—a long-term vision of creating a Bot Factory to harness the power of artificial intelligence (AI) and significantly boost operational efficiency.
From Disparate Experiments to a Standardized Framework
The rise of generative AI agents—technological marvels capable of reasoning, planning, and acting—has opened new avenues for internal productivity at AutoScout24. This prompted various engineering teams to experiment with AI, each exploring different tools and frameworks on Amazon Web Services (AWS). Recognizing the need for a coherent strategy, the AutoScout24 AI Platform Engineering team, in collaboration with the AWS Prototype and Cloud Engineering (PACE) team, held a three-week AI bootcamp. Their mission? Transition from fragmented experiments to a unified Bot Factory blueprint aimed at streamlining AI agent development across the organization.
The Challenge: Identifying a High-Impact Use Case
A key operational cost was identified: internal developer support. AutoScout24’s AI Platform engineers found themselves spending nearly 30% of their time on repetitive tasks—like answering queries, managing access, and navigating documentation. This “support tax” not only hindered productivity but also pulled skilled engineers away from critical feature development.
The solution? An automated support bot that could:
- Knowledge Retrieval: Address “how-to” questions using Retrieval Augmented Generation (RAG) techniques.
- Action Execution: Perform routine administrative tasks, such as assigning licenses through secure API integrations.
This targeted approach allowed the team to validate the blueprint while delivering immediate business value.
Architectural Overview
In order to realize the Bot Factory vision, AutoScout24 developed a robust and well-structured architecture designed for resilience and maintainability. The architecture focused on a simple, decoupled flow:
- User Interaction via Slack: Developers engage with the support bot in a dedicated Slack channel.
- Secure Ingress via Amazon API Gateway and AWS Lambda: Requests from Slack trigger AWS Lambda functions for security checks.
- Decoupling via Amazon Simple Queue Service (SQS): Verified requests are queued in an SQS FIFO queue, maintaining resilient separation between the frontend and backend.
- Agent Execution via Amazon Bedrock AgentCore: The queued messages activate the agent within AgentCore Runtime, which orchestrates calls to the AI and manages tool integrations.
This systematic design ensures that each interaction retains conversational context and builds upon previous exchanges. By tying each interaction to a unique session ID, AutoScout24 can maintain a coherent dialogue for users, transforming what could be an impersonal interaction into a more human-like conversation.
The Solution: A Reusable Blueprint Powered by AWS
The Bot Factory architecture is inherently event-driven, serverless, and relies on managed AWS services. Among its key components:
- Amazon Bedrock: Provides access to high-performing foundation models that serve as the reasoning engine for the agent.
- Amazon Bedrock Knowledge Bases: Supports RAG capabilities, enabling the bot to access internal documentation effectively.
- Amazon Bedrock AgentCore: Offers a fully managed, serverless environment for deploying and scaling the agents.
This comprehensive setup allows AutoScout24 to focus on business logic rather than underlying infrastructure. With each agent deployment running within its isolated container, data integrity and security are also enhanced.
Orchestrator Agent: Built with Strands SDK
To create the orchestrator agent, the team utilized the Strands Agents SDK, adopting a simple and declarative design. The declarative model focuses on defining agent instructions, while the foundation model deals with reasoning. The result is a clear separation of concerns that simplifies agent development.
Key snippets showcase how the agent is configured with essential tools, making it easy to add new functionalities without re-architecting the entire system.
The Impact: A Validated Blueprint for Enterprise AI
The Bot Factory project has yielded significant results well beyond the initial prototype:
- Deployment of a Functional Support Bot: The team effectively reduced the burden of routine tasks on the AI Platform Engineering team.
- Creation of a Reusable Bot Factory Blueprint: Teams can swiftly build new agents using a proven template, accelerating innovation.
- Empowerment through Abstraction: By simplifying the underlying infrastructure, a broader range of employees, including non-experts, can leverage the Bot Factory to build AI agents.
Conclusion: A New Model for Enterprise Agents
AutoScout24’s collaboration with AWS has transformed fragmented AI trials into a cohesive framework. The Bot Factory not only allowed for a smooth transition from prototype to production but also established a reusable structure conducive for future innovation. With the introduction of Amazon Bedrock AgentCore, the focus has shifted away from infrastructure, enabling a rapid and secure development of enterprise agents.
If you’re interested in building your own enterprise agents using Amazon Bedrock, numerous resources are available to guide you down this innovative path.
About the Authors
- Andrew Shved: Senior AWS Prototyping Architect guiding teams in Generative AI solutions.
- Muhammad Uzair Aslam: Technical Program Manager accelerating cloud and AI journeys.
- Arslan Mehboob: Platform Engineer with expertise in cloud infrastructure and AI technologies.
- Vadim Shiianov: Data Scientist focused on deploying machine learning and AI-driven systems.
Together, they embody the spirit of innovation, making strides in the realm of AI and cloud capabilities.