By 2028, 40% of organisations deploying AI will use dedicated AI observability tools, according to Gartner. The forecast reflects rising attention on how companies monitor model performance, bias and outputs.
Executive concern over risk management in complex AI models and agentic AI is driving the shift. Gartner defines AI observability as the use of dedicated tools to manage and assess the behaviour, decision-making and risks of AI systems, including model drift, bias and large language model logic.
The forecast points to a growing market for AI monitoring software as companies move from experimentation to production. Analysts and investors are paying closer attention as businesses look for ways to track the reliability and behaviour of models that can be difficult to explain.
The trend is also showing up in dealmaking and public markets. Cisco recently announced plans to acquire AI observability platform provider Galileo, while Datadog shares rose after better-than-expected first-quarter earnings and revenue, with growth linked in part to monitoring AI chips and coding agents.
AI observability extends beyond trust and risk to include monitoring the availability, performance and accuracy of AI platforms. Those functions become more important as organisations rely more heavily on AI-driven outputs in decision-making.
Padraig Byrne, VP Analyst at Gartner, said many companies are still at an early stage in understanding how to monitor AI systems.
"AI is everywhere, but most organisations are still figuring out how to monitor and trust these systems. That visibility gap makes scaling risky, and that's why observability matters. Unlike traditional software, AI's decision making is often hidden, making it hard to explain or trust, yet errors can cause substantial financial loss, reputational damage and regulatory scrutiny," Byrne said.
Growing demand
The move toward specialised tools is being shaped by the limits of traditional observability products, which focus mainly on infrastructure and application health. AI systems create different monitoring demands because outputs can shift over time and decision paths are often opaque.
"The shift to specialised AI observability tools is accelerating due to executive concern over risk management in complex AI models and agentic AI, not just for infrastructure or application health. There's a growing need for predictive issue detection and real-time actionable insights in AI models. Failure to adopt these tools exposes organisations to significant governance risks," Byrne said.
For infrastructure and operations teams, the lack of standardised model telemetry can lengthen incident resolution times. Engineers may have to rely on manual work to trace and debug the behaviour of deep learning models when issues arise.
"Without clear, standardised model telemetry, infrastructure and operations (I&O) teams face prolonged incident resolution times for AI applications, which will require complex manual efforts to trace and debug the behaviours of opaque deep learning models. Dedicated AI observability provides the necessary mechanisms to monitor and mitigate algorithmic risk, establishing the technical foundation for widespread enterprise AI trust and adoption," Byrne said.
Operational steps
Gartner recommends making AI model monitoring mandatory for all production deployments, with continuous tracking of fairness, drift and data quality metrics. It also advises standardising monitoring frameworks across data science, MLOps and engineering teams to reduce siloes and speed issue resolution.
The firm also urges companies to prioritise infrastructure that can ingest and analyse high volumes of model telemetry. That includes support for distributed tracing of AI inference calls, which can help teams identify where failures or performance issues emerge within AI workflows.
Another area Gartner highlighted is planning for future operational pressures as AI deployments mature. IT strategies should cover monitoring AI platform performance, detecting shadow IT activity and managing costs.
The forecast comes as companies face growing pressure to show that AI systems are reliable, explainable and well governed. For software suppliers, cloud providers and specialist monitoring vendors, that is creating an opportunity to sell tools that give operations teams more visibility into how models behave once deployed.
OpenAI and AWS are among Datadog's largest clients, underscoring the scale of demand for tools that track increasingly complex AI workloads.