In APMdigest's 2026 Observability Predictions Series, industry experts — from analysts and consultants to the top vendors — offer predictions on how Observability and related technologies will evolve and impact business in 2026. Part 9 covers Observability of AI.
AI OBSERVABILITY
In 2026, visibility will become critical for AI systems. As AI becomes a bigger piece of software architecture, the biggest worry won't just be cost and performance, it'll be trust. All those new systems look great on paper, but it only takes one high-profile goof-up before we see a lot of people suddenly much more interested in what their AI systems are doing, why they behave a certain way, and how those decisions affect systems, customers, and costs. Observability tools will need to rise to this challenge as users expect, and need, solutions that work natively with AI.
Nic Benders
Chief Technical Strategist, New Relic
In 2026, we will see increased pressure for organizations of all sizes to truly adopt and leverage AI-based technologies to realize the much-promised ROI in terms of business productivity and agility. However, AI-based insights and automation (i.e. agents) are dependent on data that accurately describes the IT infrastructure and business services that are hosted within it. Observability, therefore, moves from a useful monitoring discipline to a mission-critical capability that is fundamentally required to unlock AI-driven transformation in the modern enterprise.
Mike Nappi
Chief Product and Engineering Officer, ScienceLogic
AI has made observability essential. As teams move from experimenting to running AI in production, they're realizing how little visibility they have. You can't secure or optimize what you can't see and observability is the bridge between human judgment and machine action. In 2026, the companies that thrive will pair ambition with discipline. Resilience is the new speed. The future of software isn't human or AI, it's human plus AI, connected by observability.
Christine Yen
CEO and Co-Founder, Honeycomb
AI EUEM
Just as enterprise employee productivity extended from desktop to mobile, and the office to work from anywhere, employee productivity will extend from applications to chatbots and agentic interfaces, which will require End User Experience Management solutions to monitor AI interfaces to deliver the comprehensive visibility and resilience that enterprises require. To stay ahead, IT decision-makers need to be proactive: embed secure, enterprise-grade AI solutions into workflows, establish robust processes to audit AI usage, and educate teams on responsible practices. Endpoint management will be about governing AI-powered interactions at every user touchpoint to maintain security without stifling productivity.
Mitch Berk
Senior Director of Product Management, Omnissa
AI DEX
In 2026, digital employee experience (DEX) will be defined by "invisible AI" as copilots and agents embed themselves into workflows to summarize content, draft responses, and reduce cognitive load so employees can focus on higher-value work. However, this same shift introduces a new layer of risks as workers increasingly deploy their own shadow AI agents or use AI-powered tools without proper guidance, often exposing sensitive data to external models without realizing it. The future of DEX will be just as much about enabling workforce productivity as it is about ensuring every AI agent and AI-enabled workflow is transparent, accountable, and aligned with enterprise policy.
Mitch Berk
Senior Director of Product Management, Omnissa
DEVOPS FOR MACHINES
DevOps for Machines, Not Just Humans: DevOps is evolving beyond its traditional focus on deploying applications. DevOps for machines means governing the real-time interaction between AI agents and enterprise data, with the same rigor once reserved for production apps. Modern teams will now treat data and AI pipelines as mission-critical workloads, ensuring that AI agents have real-time, governed access to enterprise data while maintaining reliability, security, and observability at scale. DevOps for machines is about managing the data-to-action lifecycle, not model training pipelines. Humans remain responsible for defining access, policy, and safety nets. For example, tomorrow's DevOps teams will monitor not only application uptime, but also AI decision health to ensure agents operate within defined parameters. This evolution requires a new mindset: one where DevOps teams are responsible for orchestrating an ecosystem in which machines, not just humans, can operate safely, efficiently, and autonomously.
Justin Borgman
CEO and Cofounder, Starburst
AI RELIABILITY METRIC
The AI incident will become a distinct category: Organizations will start to treat AI system failures as their own incident classification, separate from traditional infrastructure or application issues. We'll see the emergence of specialized runbooks for AI model drift, hallucination events and security risks like prompt injection attacks. These incidents will require even more cross-functional than usual response teams across every part of a business, forcing a rethinking of on-call rotations and availability of subject matter experts in ML engineering, data scientists and even parts of the business that may not be used to incident response. Companies will start measuring "AI reliability" as a distinct metric alongside traditional SLOs.
Kat Gaines
Senior Manager, Developer Relations, PagerDuty
MODEL OBSERVABILITY SLO
As AI becomes just another part of the production stack, the way we think about reliability will evolve. I think we may start to see the first true "model observability SLOs," tracking things like prediction freshness and hallucination rate.
Matt Ryer
Principal Software Engineer, Grafana Labs
AUTOMATED GUARDRAILS
AI will become the biggest driver of hidden system drift because modern architectures already generate more structural change than teams can manually review. Many outages now start with small updates that no one noticed and AI will accelerate that pattern. As AI systems write code, modify schemas, and optimize configurations, the volume of change will rise faster than human oversight can scale. Engineering teams will respond by introducing automated guardrails that validate every AI action at build time before it reaches production.
Ryan McCurdy
VP, Liquibase
AI DATA OBSERVABILITY
Observability extends to AI itself: You can't optimize what you can't see, and in 2026, that includes AI models. We're already seeing this shift: organizations are bringing their AI pipelines into the same "single pane of glass" they use for applications, infrastructure, and business metrics. But as teams adopt this new generation of telemetry, they'll quickly realize that observing AI isn't actually about the model, it's about the data feeding it. Understanding the relationships between data sources, transformations, and outputs will become as critical as latency and error rates in the last generation of observability.
Matt Ryer
Principal Software Engineer, Grafana Labs
AI DRIVES COMPLEXITY
The Explosion of Apps and Agents Will Transform IT Management: Today, the average IT department manages around a hundred applications. But in 2026 that number will grow dramatically. Creating apps and AI-powered agents will become so fast and easy that IT teams could soon find themselves managing thousands of them — some running only for hours or days. This explosion will make IT environments far more complex and increase security, compliance, and data management risks. To stay ahead, organizations will need automation and intelligent tools that simplify how applications and agents are delivered, secured, and governed across any platform or cloud. The future of cybersecurity and IT management will depend on this balance between rapid innovation and strong control.
Prashant Ketkar
CTO, Parallels
Observability is all about inferring the state of applications, your classic "we don't know what we don't know" scenario. When it comes to AI, not only is the technology largely black-box in nature, but it's making ecosystems increasingly large and complex with further system, tool, and API integrations and interconnectivity. The discipline of observability will play a central role in grasping a complete understanding of enterprises' evolving systems to ensure both availability and security.
Bryan Cole
Director of Customer Engineering, Tricentis
Go to 2026 NetOps Predictions