AI-Ready infrastructure: Observability and Operations

This blind spot is exactly where most organisations find themselves today. It’s also the hidden reason as to why so many AI initiatives in operations are quietly underperforming. The bottleneck doesn’t lie with either the AI models themselves or with the budget. The true constraint is the fragmented environment the models are being forced to operate in.

The push from pilot to production is well under way, but the underlying foundations aren’t up to the task. Deloitte reports that, while 74% of organisations plan to implement agentic AI within two years, a mere 21% have the governance models necessary to managing autonomous agents safely. Capgemini echoes this warning, positioning 82% of businesses at a low or medium level of AI infrastructure maturity.

These numbers point to a single, urgent truth; intent is running dangerously ahead of readiness. The steep operational cost of that mismatch is scrutinised in AI-Ready Infrastructure: Observability and Operations, a report from Antal in collaboration with MakoLab. It makes a forceful, evidence-based case reinforcing the warning that deploying AI in an unstructured IT environment doesn’t reduce your business risk. It amplifies it.

Technical monitoring is standard practice almost everywhere. CPU load, memory usage, response times and component availability; most of that data is supplied out of the box by your cloud provider. Yet these metrics share a critical flaw in that they are incapable of telling you whether a service is actually working for your users or your bottom line.

True observability begins when an organisation monitors its critical business pathways in parallel with its IT infrastructure. In e-commerce, this means tracking the purchasing journey in its entirety. In an enterprise system, it means safeguarding the operations that business continuity, customer service and revenue depend on.

As yet, very few organisations have made this leap successfully. The collaborative report is remarkably direct about why they’re stalling and the steep financial price they’re paying for remaining in the dark.

Settling for reactive observability might seem easier to justify in the short term. It is, however, a false economy that builds zero resilience.

When organisations turn to a mature approach, the data tell a different story. AI-Ready Infrastructure identifies five critical decision-making categories that improve measurably when observability goes proactive. A brief outline of two of them follows:

·     targeted resilience investment: instead of applying redundancy just because a system is easy to instrument, high-availability mechanisms can be deployed at exactly the point where a failure would cause the greatest commercial damage;

·     derisked releases and changes: your business gains the ability to forecast the operational impact of planned releases and configuration changes accurately, before they ever touch your production environment.

The details of these and the other three strategic advantages can be found in the report. Crucially, it also reveals the fundamental shift in how incident costs should be calculated. This is a new framework that underpins all five of the proactive decisions.

The sharpest argument in the report, this deserves stating plainly. Artificial intelligence is only as useful as the ground truth it works from. When documentation is inconsistent, runbooks are out of date, incident history is patchy and ownership is undefined, an AI model has no reliable context. It will still generate an answer and that answer will sound incredibly confident. However, it most certainly won’t be a dependable basis for a critical operational decision.

This brings us to the very question the report was written to answer. What, specifically, should your IT environment contain for AI to become a genuine asset, rather than an accelerant for existing chaos?

The answer has far less to do with software tooling than most teams expect. AI-Ready Infrastructure maps out the requisite operational knowledge model in full, revealing exactly which inputs matter, what ‘consistent, up-to-date and relationally connected’ actually means in practice and the level of automated resilience this groundwork unlocks on the other side.

The upside of AI in operations is well documented. Teleport records clear improvements in incident analysis times, documentation quality and engineering efficiency for organisations leveraging AI.

On the other hand, the exposure is equally real. Three in five organisations have either already experienced an AI-related security incident or strongly suspect that they’ve been subject to one. An astonishing 70% of teams admit that they grant their AI systems broader access than a human performing precisely the same role.

The clearest warning signal here concerns permissions. Organisations running over-privileged AI systems suffer a staggering 76% incident rate. The report details the corresponding figure for environments where strict ‘least privilege’ access is enforced. Interestingly, the gulf between the two is far wider than most engineering teams assume.

Handing over full autonomy in a production environment continues to raise legitimate alarm about transparency, traceability and auditability. Ultimately, it comes down to one undeniable fact. When things go wrong, accountability lies squarely with your people and your organisation, not with the model.

The question isn’t whether to draw a line at all, but where exactly it needs to go.

AI-Ready Infrastructure takes a firm stance on this. It provides the outline of a practical framework for dividing the workload, indicating which tasks should be handled by technology, which critical decisions it is imperative to leave with humans and the precise allocation of responsibility when technology and people are working in tandem.

The report also covers another vital point, detailing the seven defining characteristics shared by organisations that are genuinely AI-ready. None are exotic. Most are entirely unglamorous. All of them, though, are infinitely harder to retrofit than they are to build from the start.

• The complete observability maturity argument, Including a practical diagnostic tool for benchmarking your current systems accurately.

• The operational knowledge model revealing not only which data inputs transform AI into a useful asset for incident work, but also the capabilities they unlock.

• A robust human-in-the-loop framework providing a clear guide that maps precisely where machine automation should end and human accountability should begin.

• The 7 defining characteristics of an organisation that is genuinely ready to deploy AI safely in production operations.

•  Exclusive, trench-level insights and commentary from MakoLab’s own cloud, infrastructure and DevOps experts.

• Fully transparent data, with every external figure rigorously sourced and dated, offering you reliable evidence for building your own internal business case.

AI-Ready Infrastructure provides the blueprint, but the most valuable conversation is the one that gets personal, zooming in on your own architecture.

Book a dedicated working session with our Cloud & Infrastructure Services team. They will map the report’s AI-readiness criteria directly onto your environment, evaluating your critical services, system ownership, telemetry coverage, runbook currency and access models.

Together with you, they will pinpoint exactly which operational gaps expose you to the greatest business risk and outline the steps that need taking to close them.

AI-Ready Infrastructure: Observability and operations is the inaugural publication in an extensive research series by Antal in partnership with MakoLab.

Designed for IT leaders and decision-makers, the complete programme maps the future of enterprise technology. Upcoming releases will delve deeply into:

·     next-generation software delivery through agentic engineering;

·     AI governance and data sovereignty;

·     the practical applications of neuro-symbolic AI;

·     AI adoption and readiness across European enterprises; a comprehensive quantitative study tracking the current state of play.