C.J. Oosthuizen – Google Cloud & Workspace Specialist at ATG, part of global entity Smollan, explains: “Six months ago, a wave of optimism swept through enterprise IT. Organizations were spinning up Large Language Models (LLMs), piloting productivity assistants, and rushing to claim their stake in the AI revolution. At the time, 40% of IT leaders confidently declared their AI deployments to be ‘mature’. Today, that confidence has collapsed,” he says.

Oosthuizen notes the latest, published quarterly,  IT Trends report from JumpCloud for Q3, highlights how IT teams are actually adopting AI with the number of leaders who consider their AI deployments mature has plummeted from 40% to just 23%. “This drop isn't because AI adoption has slowed down. In fact, it is the exact opposite. What we are witnessing is a striking paradox in that more AI is in active use than ever before, yet there is significantly less confidence in how it is being implemented,” says Oosthuizen.

He says that hard truth has caught up with hype. “As AI agents gain deeper access to enterprise data and execute tasks with increasing autonomy, business people are realising that their underlying governance and infrastructure have not kept pace.”

He explains the numbers reveal high autonomy and zero visibility. “The data from this report paints a stark picture of an environment moving faster than the guardrails built to protect it. The Silicon Workforce isn't coming, it’s already here, and running at a pace that is outstripping traditional IT controls.”

Oosthuizen synopsises some of the findings as follows:

Agents Are in Production, and They’re Flying Solo

• The shift from passive, text-generating chatbots to active, workflow-executing agents has happened almost overnight. More than 60% of organisations are reported to be running AI agents in production workflows.

“Worse yet, we are removing the human safety net. Full AI autonomy without human review has more than doubled in just a short window, jumping from 11% to 26%. More than a quarter of enterprise AI implementations are now making decisions, manipulating data, and executing tasks entirely on their own.”

The Explosion of Non-Human Identities

• This rapid agent deployment has fundamentally altered the perimeter of enterprise networks. In 83% of organisations, non-human identities (NHIs) - such as API keys, service accounts, bots, and AI agents - now outnumber human users. Yet, despite machine identities dominating the network, only 21% of organisations have established governance controls for them.

“This means the vast majority of the identities operating inside enterprise systems are completely unmonitored, unmanaged, and unaccountable,” Oosthuizen emphasises.

The Risk Velocity Problem

“When you combine autonomous execution with a total lack of identity governance, risk escalates exponentially. It’s no surprise, then, that 3 in 4 IT leaders (75%) are reported to admit that AI is advancing faster than their ability to manage the associated risks,” he confirms.

Why the Confidence Bubble Burst

The report notes the sudden drop in perceived AI maturity is actually a sign of healthy realism. “Six months ago, "maturity" meant successfully deploying an API or launching an internal corporate chatbot. Today, IT leaders understand that true maturity means lifecycle management, data lineage, continuous compliance, and identity security. They are realising that the traditional IT toolkit is fundamentally unequipped for this new paradigm.

“When a human employee leaves an organisation, deprovisioning their access is a standard workflow. But when an autonomous AI agent is hardcoded into three different production systems via a legacy service account, the question is how do you revoke its permissions without crashing the business? Most organisations don't have an answer to that question. That is why confidence is plummeting.”

Reclaiming Control: The Path Forward

According to the report the solution to the AI confidence gap isn't to stop innovating or to ban AI tools, user-led demand will always find a way around a firewall. Instead, IT leaders must rapidly shift their focus from the AI applications themselves to the identity and infrastructure layers that support them.

It is recommended that to safely scale the Silicon Workforce, enterprise infrastructure must evolve across three critical areas:

• Unify Human and Machine Identity: If machine identities outnumber human users, they can no longer be treated as secondary configuration details. Non-human identities must be brought into the central directory, subjected to the same strict Zero Trust access controls, and continuously monitored.
• Implement Just-In-Time (JIT) Privileges for AI: Autonomous agents should not possess permanent, blanket administrative access to databases. Infrastructure must adapt to grant temporary, tightly scoped permissions that expire the millisecond a specific workflow is executed.
• Bridge the Visibility Gap: IT teams require a single control plane capable of identifying every AI agent, understanding what data it has access to, and tracking its autonomous actions in real-time.

Oosthuizen concludes: “The initial rush of the AI boom was about what the technology could do. This next phase is about what the technology should be allowed to do. By shifting investment away from the shiny objects to the unified, identity-centric infrastructure, IT leaders can close the risk gap and turn a volatile environment back into a competitive advantage.”