As AI systems move rapidly from experimentation into production, organizations are discovering that adoption alone is not the hard part, understanding, governing, and trusting AI in live environments is. 

In this episode of the Tech Transformed, Shubhangi Dua speaks with Camden Swita, Head of AI, New Relic, about why AI observability has become a critical requirement for modern enterprises, particularly as agentic AI and AI-driven operations take on increasingly autonomous roles.

The discussion explores how traditional observability models fall short when applied to probabilistic systems, why many AI ops initiatives stall at proof-of-concept, and what security and IT leaders must prioritize to safely scale AI in production.

Be the first to see how intelligent observability takes you beyond dashboards to agentic AI with business impact at New Relic Advance, February 24, 2026.

Why AI Adoption Is Outpacing Operational Readiness

While AI adoption is accelerating rapidly, most organizations still lack visibility into what their AI systems are actually doing once deployed. Generative AI is already widely used for natural language querying, coding assistants, customer support bots, and increasingly within IT operations and SRE workflows. 

As these systems move into production, new challenges emerge around cost control, governance, performance quality, and trust. Leaders recognize AI’s potential value, but without deep observability, they struggle to determine whether AI-enabled systems are delivering consistent outcomes or introducing hidden operational and security risks.

How Observability Must Evolve for Agentic AI and AI Ops

The episode then examines how observability itself must evolve to support agentic and autonomous AI systems. While core observability principles still apply, AI introduces a new layer of complexity that requires visibility into model behavior, agent decision-making, and multi-step workflows. 

Modern AI observability extends traditional application performance monitoring by capturing telemetry from LLM interactions, agent orchestration layers, and automated evaluations of output quality against intended use cases. 

Without this visibility, teams are effectively operating blind, unable to diagnose failures, validate compliance, or confidently deploy AI at scale. At the same time, AI is increasingly being embedded into observability platforms to reduce noise, accelerate root cause analysis, and improve incident response.

Making Agentic AI Work in Practice  

Successful adoption starts with low-risk, high-friction tasks such as incident triage, dashboard interpretation, and runbook summarization, rather than fully autonomous remediation. These use cases deliver immediate productivity gains while preserving human oversight. Over time, stronger feedback loops, better context management, and human-in-the-loop learning allow agents to become more reliable and useful. Looking ahead, Camden predicts that 2026 will be a turning point for agentic AI in production, driven by maturing AI observability platforms, richer semantic data, and knowledge graphs that connect technical telemetry to real business outcomes.

Key Takeaways

AI adoption is accelerating in enterprise environments.

Organizations face complexities in productionizing AI features.

Natural language querying is a common AI application.

AI agents are increasingly used in IT operations.

Observability is crucial for understanding AI systems.

Traditional observability solutions are evolving to include AI monitoring.

Incident response teams struggle with alert noise and context gathering.

AI can assist in incident management and root cause analysis.

Future trends include more reliable AI agents and monitoring solutions.

Organizations need to invest in AI observability to succeed.

Chapters

01:20 The Current State of AI Adoption

02:28 Purposeful AI Usage in Organizations

04:40 Observability in the Age of AI

08:05 Evolving Observability Solutions

11:36 Challenges in Incident Response

16:04 Integrating AI in Operations

23:33 Future Trends in AI Monitoring

30:29 Investment Strategies for AI Solutions