Yrieix Garnier, VP of Product at Datadog, frames the current shift to AI as an extension of trends that began with cloud migration and digital transformation. In his view, organizations are already operating in highly fragmented environments, with increasing numbers of teams, tools, and services contributing to system complexity. AI accelerates this trend by introducing more models, faster development cycles, and additional layers of interaction across systems. The challenge is not simply adopting AI, but maintaining control and visibility as complexity scales.

Garnier positions observability as the core mechanism for managing this complexity. Rather than treating AI systems as isolated components, he emphasizes the need to connect infrastructure, services, and teams into a unified view. This includes correlating telemetry across logs, metrics, and traces to understand how systems behave under load and during failure conditions. As AI systems are introduced, this same approach extends to monitoring model behavior, agent interactions, and system-wide dependencies. The goal is to create a continuous understanding of how distributed systems operate in real time.

A central part of his perspective is the role of AI agents in operations. Garnier describes a shift from fragmented, manual incident response toward systems that can automatically detect issues, generate hypotheses, and investigate root causes across multiple data sources. These agents interact with logs, traces, codebases, and external tools to build a structured understanding of system behavior. 

This shift introduces new requirements around explainability and control. Garnier highlights the importance of being able to trace how an AI system arrives at a conclusion, including the sequence of actions it takes and the data it accesses. Features such as agent traces and structured telemetry are designed to make these processes visible and auditable. In environments where AI systems participate directly in operational workflows, the ability to understand and validate their behavior becomes a prerequisite for deployment.

He also emphasizes that AI observability must extend across the full system stack. This includes infrastructure-level insights such as GPU utilization and cost, model-level performance and input-output behavior, and application-level interactions between agents and services. By integrating these layers into a single platform, organizations can reduce fragmentation and respond more effectively to issues that cross system boundaries.

Another implication of Garnier’s approach is the integration of observability into development workflows. By allowing developers and AI agents to interact directly with live system data, the gap between detection and remediation can be reduced. Systems can surface issues, suggest fixes, and even generate changes within development environments, creating a tighter feedback loop between production and engineering teams. 

As AI becomes embedded across workflows and services, the ability to manage complexity through observability and coordinated control systems becomes the defining factor in how effectively organizations can deploy and operate AI in production.

Enterprise AI agents depend on structured governance. Because they operate within a complex ecosystem of data and tools, they require rigorous access management and guardrails. Modern agentic systems rely on multi-agent orchestration, and this constant interaction increases the need for real-time monitoring and policy enforcement. Ultimately, successful deployment is about the underlying architecture that provides centralized control across all agents and workflows.

This enterprise AI conference brings together executives, engineering leaders, infrastructure teams, and product strategists building AI systems. The theme is Orchestrating AI Autonomy, focused on connecting agents, data, and infrastructure into scalable, governed setups with clear ROI. Speakers include leaders from Instacart, Asana, Mastercard, Intuit, eBay, LangChain, Amazon, CrewAI, Runway, and Stanford.  Details →

This massive industry gathering is the largest meeting of enterprise AI leaders, bridging the gap between technical researchers and business decision-makers. The event covers the entire AI ecosystem, focusing on use cases across 30+ industries, from financial services and healthcare to retail and energy. Speakers include executives from top-tier organizations such as Google, Netflix, Meta, and Amazon, as well as various Fortune 500 companies. Details →

This report examines the growing risks associated with deploying AI systems, particularly as agents become more autonomous in real-world environments. It highlights that current systems still exhibit reliability issues, including incorrect outputs, flawed reasoning, and unsafe actions, and that these risks are amplified when agents operate with greater independence across tasks and systems. The findings emphasize that as agents gain more autonomy, it becomes harder for humans to intervene before failures occur. While existing safeguards can reduce error rates, they remain insufficient for many high-stakes applications. Read →

In this episode, Waxell AI CEO Logan Kelly explains why a dedicated governance and control layer is the essential missing piece in the modern AI agent stack. While most founders prioritize functionality and automation, Kelly argues that deploying agents without proactive guardrails exposes companies to significant risks, including context-poisoning attacks and runaway API costs. By shifting the focus from simple observability to active governance, leaders can move beyond basic chatbots to scalable agentic infrastructure that remains secure, predictable, and aligned with human oversight.