AWS re:Invent 2025 - Agents in the enterprise: Best practices with Amazon Bedrock AgentCore(AIM3310)

Source: Dev.to

Overview

AWS re:Invent 2025 – Agents in the enterprise: Best practices with Amazon Bedrock AgentCore (AIM3310)

In this session, AWS Principal Product Manager Kosti Vasilakakis and Tech Lead Maira Ladeira Tanke present best practices for moving agentic AI from proof‑of‑concept (POC) to production using AgentCore. They outline nine key rules, including:

• Start small with defined use cases
• Implement observability from day 1 (OpenTelemetry)
• Expose tools with clear descriptions
• Adopt multi‑agent architectures
• Scale securely with user‑specific memory and identity controls
• Use deterministic code for calculations
• Continuously test with evaluations

Live demos show how to build agents with frameworks such as Strands and LangChain, integrate Gateway for API access, implement a Code Interpreter for analysis, and deploy with Runtime’s micro‑VM isolation.

Phil Norton from Clearwater Analytics shares their experience building 800 agents and 500 tools, emphasizing “context is king,” and describes their migration to AgentCore for zero‑downtime deployments and elimination of noisy‑neighbor issues in financial data processing workflows.

Watch the session on YouTube

Introduction: Bridging the Gap Between Agent Demos and Production Scale

“If you’ve ever seen an agent demo that worked perfectly and then wondered how to scale it securely for many users while maintaining performance, this is the room for you.”

The session is designed for three personas:

1. Developers / Engineers building agents
2. Platform teams providing the underlying infrastructure
3. Business stakeholders who use or support the agents

Speakers

• Kosti Vasilakakis – Principal Product Manager, Agentic AI
• Maira Ladeira Tanke – Tech Lead, Agentic AI

Later, Phil Norton, Senior Manager of Software Development at Clearwater Analytics (a unified investment platform with > 10 trillion AUM), joins to discuss real‑world production challenges.

The POC‑to‑Production Chasm: Six Critical Capabilities for Scaling Agents

Customers repeatedly tell us that moving from a POC to a production‑ready agent is difficult. The gap can be bridged by mastering six capabilities:

1. Accuracy – Agents must perform reliably with real users, whose behavior often differs from developer expectations. Maintaining high accuracy is essential as user expectations evolve.
2. Scalability – Agents need to serve many users across diverse domains while preserving hyper‑personalization. This includes handling per‑user memory securely.
3. Security – Production agents access live systems and real data. Robust identity controls and secure memory handling are non‑negotiable.
4. Cost Management – Token usage and hosting infrastructure drive expenses. Visibility into cost drivers enables effective budgeting and optimization.
5. Observability – An end‑to‑end observability pipeline (metrics, logs, traces) is required to understand agent behavior, diagnose issues, and drive continuous improvement.
6. Monitoring – Ongoing monitoring of accuracy, latency, cost, and security ensures agents remain reliable as underlying open‑source components evolve rapidly.

“How can we embrace open source securely, scalably, and with full observability?” – a question many customers ask, and the primary motivation behind AgentCore.