How Mitalee Gujar Made Amazon’s Large-Scale Infrastructure Migration Safer with AI

Episode Summary:

In this episode of Engineering Choices You Have to Defend, hosted by Nicola Onassis, we sit down with Mitalee Gujar, Director of Engineering at Amazon, to explore the challenges of modernizing large-scale infrastructure while protecting critical business operations. Drawing on her experience at Amazon, Google, and Cisco, Mitalee shares how she is leading a major networking and connectivity transformation at Amazon, consolidating legacy systems into a more secure, scalable, cloud-native architecture designed to support the company's evolving needs and the future of agentic AI.

Mitalee explains why moving faster is not always the best approach to a complex infrastructure migration. With thousands of services and millions of connections supporting critical e-commerce operations, even a small change can trigger widespread outages. She discusses how legacy dependencies, limited visibility across interconnected services, and noisy monitoring alerts made the migration particularly challenging. By investing in observability, improving alert accuracy, and planning rollback strategies before deployment, her team worked to reduce false alarms from approximately 60% to 5% and make migrations safer.

The conversation also explores how AI is transforming infrastructure engineering. Mitalee shares how her team applied AI and agentic frameworks to automate onboarding support, answer technical questions, improve troubleshooting, and reduce support ticket volume by approximately 30%. She also discusses AI's potential to modernize legacy code, automate testing, and simplify complex infrastructure changes. While AI can accelerate engineering work, she emphasizes the importance of proper evaluation, safeguards, and human oversight. Her central lesson for engineering leaders is that resilience, reliable monitoring, and proactive communication are essential investments before accelerating any high-risk migration.

Key Takeaways:

• Prioritize resilience before speed: Invest in monitoring, alerting, and recovery strategies before accelerating high-risk infrastructure migrations.
• Understand service dependencies: Mapping interconnected services and their clients helps teams identify risks that infrastructure-level monitoring alone may miss.
• Separate critical signals from noise: Better alerting systems help engineering teams identify genuine regressions and make informed rollback decisions.
• Use AI to reduce operational overhead: AI-powered knowledge bases and agentic frameworks can streamline onboarding, automate support responses, and improve troubleshooting.
• Modernize legacy systems with AI: AI can help transform outdated infrastructure, improve efficiency, and support business growth when paired with appropriate safeguards.
• Plan rollbacks before deployment: Define risk levels, migration milestones, monitoring requirements, and recovery options before making changes.
• Coordinate across teams: Proactive communication with dependent services and support teams helps detect issues early and minimize business disruption.
• Keep humans in control: AI can accelerate infrastructure engineering, but careful evaluation, guardrails, and human oversight remain essential.

Connect with Mitalee Gujar:

• LinkedIn: linkedin.com/in/mitalee-gujar
• AI Infra Summit Speaker Profile: ai-infra-summit.com/speakers/mitalee-gujar

Listen Now & Subscribe:

Apple Podcasts, Spotify, Amazon Music, or wherever you get your podcasts.

"Engineering Choices You Have to Defend explores the real technical decisions behind regulated software, engineering transformation, AI integration, and the systems leaders build to make complex technology work in the real world."