"A productive equilibrium": How to scale agentic AI without losing control
"AI agents are emerging not as a replacement for human capability, but as powerful partners in amplifying it."
Over the past year, the conversation around AI has matured considerably. Rather than asking what’s possible, organisations are increasingly focused on what it takes to deploy agentic AI in ways that are secure, governed and capable of delivering long-term business value.
After an initial wave of experimentation, many organisations now find themselves at a crossroads. Proof of concepts have demonstrated potential, but the next challenge is embedding agentic AI into day-to-day operations without introducing unnecessary risk or complexity.
Our latest research into agentic AI reflects this shift in priorities. Business leaders are no longer asking whether autonomous AI has a role to play, but how they can deploy it responsibly. Questions around governance, oversight and trust have overtaken discussions about model capabilities, as organisations look to move confidently from isolated pilots to production environments.
Encouragingly, investment intentions remain strong, with nearly three-quarters of organisations planning to increase spending on agentic AI over the coming year. However, bigger budgets alone won’t unlock transformation. Unless organisations overcome the barriers preventing AI from scaling, they risk creating more successful pilots rather than meaningful business outcomes.
Across EMEA, those barriers are becoming increasingly clear. While access to skills remains important, security and data privacy concerns continue to be the biggest obstacles, with more than half of organisations identifying them as the primary challenge to wider adoption.
Why successful pilots don’t automatically lead to successful AI adoption
It’s clear that agentic AI projects won’t deliver substantial value if they remain stuck in the pilot phase.
So far, most deployments have concentrated on IT operations, but we’re beginning to see that change. More organisations are applying agentic AI in repeatable, customer-facing areas like customer support, where the impact is even more visible. Even in fields like legal services - currently one of the slowest sectors to deploy AI - automation is expected to grow significantly over the next few years, showing that confidence in more complicated use cases is building.
Interestingly, the same capabilities that can streamline IT workflows can also reshape how businesses serve their customers and drive commercial growth. Yet, as AI moves closer to customers and those core decision-making processes, the stakes naturally get higher, requiring greater oversight and guardrails to be put in place.
Trust, governance and the role of human oversight
For today’s organisations, safeguarding security and privacy are top criteria for moving projects from pilot to production. However, for many leaders, the real barrier is not the complexity of the technology itself, but trust.
Establishing trust and confidence means identifying clear boundaries for when an AI agent can act autonomously but also guaranteeing human oversight at critical decision points. It’s increasingly clear that in the era of agentic AI, trust has become the ultimate control mechanism.
As it stands, nearly 70% of agentic AI decisions are currently verified by humans, and almost half of organisations conduct a human-led review of AI outputs as a verification measure. This signals a deliberate balance: the pendulum is not swinging entirely towards automation, nor is it reverting back to purely human control.
READ MORE: Agentic AI demands an upgrade to financial system resilience, Bank of England warns
Instead, we’re seeing a productive equilibrium. Human judgement and agentic AI are complementing one another - AI executes with speed, while humans provide direction and guardrails. Agentic AI is emerging not as a replacement for human capability, but as a powerful partner in amplifying it.
As organisations integrate agentic AI more deeply into workflows, leaders must understand and apply this division of responsibility. AI may perform the execution, but humans must continue to define goals, set boundaries, and critically, retain accountability.
Why observability underpins trustworthy autonomous AI
Business observability is fundamentally what makes this human-AI partnership sustainable, ensuring traceability and confidence at the human-AI interface.
As agentic systems grow more autonomous and interconnected, their complexity also increases. A small error in one model component - a hallucinated output or a misinterpreted prompt - can quickly cascade across applications and environments. With a significant number of teams still manually reviewing agentic AI communication flows, this reveals a critical gap in real-time, context-aware automation.
Without comprehensive visibility, organisations are forced into a reactive stance - diagnosing issues only after they’ve already introduced risk. Simply logging events or flagging anomalies after the fact is no longer sufficient. Today’s organisations need systems that can detect hallucinations and anticipate downstream impacts in real time, before they escalate into material problems. In increasingly complex multi-model and multi-agent ecosystems, observability is therefore the backbone of scalable, trustworthy autonomous operations.
From experimentation to enterprise-wide adoption
The enterprises that realise the greatest value from agentic AI won’t necessarily be those running the highest number of pilots. They’ll be the ones that create the right foundations to deploy AI safely, transparently and at scale.
That foundation is built on trust. By combining robust governance, human oversight and deep observability, organisations can give autonomous AI the context and guardrails it needs to operate with confidence.
READ MORE: "Nine seconds of terror": How to stop AI agents wiping out systems at machine speed
Rather than slowing innovation, these capabilities enable organisations to expand AI into increasingly critical business processes while maintaining control.
Ultimately, the next phase of AI adoption isn’t about proving the technology works. It’s about proving organisations can rely on it every day. Those that engineer trust into their AI strategies from the outset will be best placed to turn early experimentation into lasting operational and commercial advantage.
Joshua Clay is RVP Solutions Engineering at Dynatrace UK&I
