the voice of customer experience technology
Front pagesponsored · AWS (Amazon Web Services)
3 Dec 2025 · 2 min read

The Flight Simulator for Agentic AI: Navigating Observability in Amazon Connect

AWS re:Invent 2025: Mike Wallace explains how Amazon Connect acts as a "Flight Simulator" for Agentic AI—enabling safe testing, observability, and guardrails.

As businesses rush to deploy autonomous AI agents, the fear of the "black box" remains the biggest barrier to entry. At AWS re:Invent 2025, Amazon Connect’s Americas Solutions Architecture Leader Mike Wallace reveals how to move from hype to production without flying blind. 

In the race to adopt Agentic AI systems that can reason, make decisions, and take action on behalf of customers, CX leaders are facing a critical safety challenge. How do you trust an AI to negotiate refunds or handle sensitive data without a human in the loop? 

According to Mike Wallace, the answer lies in treating your AI exactly like a pilot. 

Take that same care of your Agentic AI... apply the same logic that you put against your human agents into your agent AI, and you'll be successful

In this exclusive interview from the re:Invent expo floor, Wallace outlines Amazon Connect’s new architectural approach to AI Observability. He argues that before an agent ever touches a live customer, it must pass through a rigorous "Flight Simulator"—stress-testing thousands of scenarios to ensure it doesn't crash when turbulence hits. 

Watch the full interview to uncover: 

  • The "Flight Recorder" for Reasoning: How to see not just what the AI said, but the chain of thought it used to get there. 
  • Pre-Flight Stress Testing: Wallace explains how to run thousands of automated simulations to catch "hallucinations" before they reach production. 
  • The "Instructor Override": How Contact Lens now acts as a real-time co-pilot, detecting customer frustration or policy breaches and handing off to a human supervisor instantly. 
  • Bounding the Airspace: A look at how Guardrails for Amazon Bedrock can block competitors' names and enforce brand safety at the architectural level. 
You can't just put your API at a Large Language Model and hope for the best

Wallace warns. 

If you are planning to deploy Agentic AI in 2025, this conversation provides the blueprint for ensuring your agents stay on the runway and out of restricted airspace. 


For our all our event coverage, announcements and excusive interviews visit our AWS re:Invent news hub here. 

The discussion0 takes · attributed & checked

Does this reflect your experience?

opening the room…
Read nextordered by techtelligence · every pick explained
more from AWS (Amazon Web Services) · sponsored

Why AI Agents Must Be Proven Before They Are Deployed

26 May 2026
matches this topicVerint Warns AI Value Needs Better Data3 Sept 2026matches this topicSalesforce’s Outcome-Based Pricing: Strategic Masterstroke or Defensive Scramble Against AI Seat Compression?27 Aug 2026