BLP's agentic AI platform resolves millions of exceptions every day across hundreds of enterprise customers in more than 40 countries. copebit builds and operates the AWS foundation that powers it. This talk explores both sides of the equation: the agentic AI platform and the infrastructure that makes it enterprise-ready.
First, the production system. Rather than relying on document retrieval, BLP mirrors each customer’s ERP environment — including master data, business rules, and process logic — in the BLP Digital Twin. More than 300 pre-trained AI agents operate against this structured model through 30+ standard ERP integrations. The most appropriate AI model is selected for each processing step through Amazon Bedrock, leveraging Anthropic’s Claude Sonnet and Claude Opus models alongside Gemma, with all inference running in European AWS regions.
Then, the lessons for your own implementation. Drawing on our experience building AWS foundations for Swiss enterprises and operating agentic AI platforms on Amazon Bedrock and Amazon Bedrock AgentCore, we explore four architectural decisions that determine whether an enterprise agent platform can withstand a security review:
How user identity and permissions propagate through to target systems.
How to manage the size of the tool catalog without compromising context efficiency or cost.
How to ensure endpoint stability as the platform evolves.
How to make the registration of new tools a governed, reviewed change.
For each decision, we share the architectural trade-offs, practical lessons, and failure modes encountered along the way.
Attendees will leave with a real-world perspective on what it takes to build, secure, and operate agentic AI at enterprise scale.
Put your tools, products, and engineering work in front of builders working with agentic AI.