Скачать или смотреть Solving the IAM "Black Box" and Building Agents with Long-Term Memory

In this week's discussion (Jan 21, 2026), we dive into the latest AWS and GenAI developments that caught our attention. From significant IAM troubleshooting improvements to the evolution of AI agent "wisdom," we break down the technical nuances you need to know.

In this episode:
IAM Visibility Upgrade: AWS now includes policy ARNs in Access Denied messages—no more hunting through SCPs and permission boundaries blindly.
Bedrock Episodic Memory: A deep dive into how agents now capture reasoning chains and "reflections" to improve future performance.
The Strands SDK: Exploring AWS's model-driven agent framework and the pitfalls of AI-generated "vibe coding."
CloudWatch SQL Alarms: How to monitor entire EC2 fleets using dynamic SQL queries instead of static resource IDs.
Token Efficiency Benchmarks: Why your choice of programming language might be eating your LLM's context window.

Timestamps:
01:45 - IAM: Policy ARN Visibility in Error Messages
03:55 - Bedrock: Episodic Memory & Reflections
10:45 - Strands SDK & Multi-Agent Collaboration
14:30 - The "Vibe Coding" Case Study
24:30 - CloudWatch: Fleet Management with SQL Alarms
49:30 - Token Efficiency: Benchmarking 2026 Languages
01:01:00 - Weekly Media: The Rip, Osiris, and Hightail