AI agents are entering IT teams, but one missing layer could decide who scales fastest

Enthusiasm for AI adoption is at an all-time high, and rightly so. AI is transforming how software is created, delivered, and operated, unlocking significant productivity gains. While some organizations use generative AI to enhance existing workflows, others are taking an AI-first approach with agentic AI to reimagine software development and operations. As adoption expands, executives increasingly see AI-powered observability as the bridge from human-driven operations to human-supervised autonomous systems.

The generative AI everyone is talking about is a probabilistic system - a creative artist brilliant for brainstorming, generating novel content, and accelerating ideation. But critical IT operations need a scientist: a deterministic, fact-based system that is reliable and comprehensible. Just as a calculator is deterministic. This scientist reasons from verified data such as real-time topology, causal dependencies and precise metrics to deliver answers teams can act on with confidence.

This is where the promise of Agentic AI - systems that can independently reason and act - needs the right foundation. While 50% of organizations have agents in production for limited use cases, only 23% have scaled these projects to mature, enterprise-wide integration. The gap signals a maturing ecosystem, not AI failure - and those who solve for context, accuracy, and feedback loops will scale fastest.

One of the key challenges is that generative AI can hallucinate and push agents off course. CTOs rightly prioritize ensuring that agentic systems have instant access to high-quality information, so multi-step agent workflows can execute quickly and reliably. Hallucinations aren't minor errors, they can trigger wrong actions leading to outages and security risks. In agentic chains, inaccuracies can accumulate and amplify, resulting in financial exposure.

But these are solvable problems. The combination of contextual observability, deterministic AI, and real-time dependency mapping is already making agentic AI more reliable and actionable.

So, how do we move forward? By establishing an AI ecosystem that encompasses AI-powered observability for reliable results. Organizations start with clearly defined use cases, implement strong oversight, and gradually expand AI's role as confidence in its outputs grows. The key is to combine the creative power of generative AI with a deterministic foundation grounded in the hard facts of operational data. This creates a system where humans set strategic goals while reliable AI handles tactical execution with precision, guided by established policies and guardrails.

The journey from automation to autonomy is an evolution. Each organization progresses through its own maturity stages on the journey to autonomous operations. For the majority, it takes shape across three distinct stages.

The first stage is automated. The journey begins by moving beyond simple, brittle scripts. In this stage, the system performs well-defined tasks based on AI-generated answers rooted in real-time, contextual data. It's about reliably automating responses to known problems. Many organizations are striving to reach this stage or are already in it, and that progress is accelerating.

The second stage is supervised autonomous. As trust in the system grows, it graduates to handling more complex scenarios. AI can analyze a novel situation, understand its business impact, and generate a ready-to-implement action plan. However, this plan is not executed until a human expert gives their approval. This keeps humans in the loop for critical decisions while offloading the cognitive burden of the initial analysis. Key principles here are reliability, transparency, and a precise feedback loop.

The final stage is fully autonomous systems which operate independently to achieve business goals - dynamically managing environments, optimizing costs and performance, and remediating issues before they impact users. The system continuously observes itself to self-optimize, ensure compliance, and provide insights that help people refine their goals. People still play a crucial role: they review outcomes, adjust strategies, and set direction. Think of the human role as an entrepreneurial-minded architect working alongside AI — focused on knowledge management, defining goals, and shaping what the system should deliver. As a result, organizations deliver and operate software with higher resilience, happier customers, and lower cost.

AI agents are powerful — they can code faster, refactor at scale, and process information beyond human capacity. But AI has no awareness of what's happening in production. It's blind to the real world without observability feeding it context.

This is why observability is mandatory for reliable AI. And this is where the real opportunity lies — not in collecting more data, but in making that data immediately usable for precise decisions.

Organizations need systems built on a unified AI data lakehouse that continuously ingests and structures telemetry, combined with a real-time dependency graph that maps every service, transaction, and infrastructure component in context. Such a foundation allows AI to be reliable and capable — providing the memory that AI needs, at scale.

This precision is critical because LLMs cannot directly process petabytes of heterogeneous observability data - context windows are limited, and performance degrades as input grows. Curating the most relevant information beats providing everything at once.

To overcome these constraints, it's essential to rapidly distill vast data into short, high-quality context. What AI agents need most is crisp, accurate context delivered at speed - where contextual analytics, dependency graphs, and an AI-optimized data lakehouse become critical differentiators.

Building this future is not just an innovation; it's a business necessity. Spending on AI-optimized infrastructure to power these systems reached $82 billion in a single quarter in 2025 and is projected to hit $758 billion annually by 2029. The rewards for getting this right are immense.

The fusion of deterministic and agentic AI represents the next chapter for enterprise IT, creating systems where AI observes and manages AI-driven environments to deliver greater digital resilience and better customer experiences. As this transformation accelerates, observability is evolving into a strategic capability that enables organizations to confidently adopt AI-first operations and build a truly autonomous future.