AI agents have come a long way from concept to a practical tool, driven by a surge in real-world use cases and accelerating enterprise demand. By 2028, 33% of enterprise software applications will include agentic AI, up from less than 1% of enterprises in 2024. Today, production-ready agents are already supporting coding assistance, procurement workflows and financial analysis across industries. Despite rapid progress in the technology, many enterprises still struggle to realise AI’s full value at scale. Scaling often comes with difficult trade-offs, as organisations try to balance cost, quality and speed. So, what’s really holding them back?
The answer lies not in the technology itself, but in the organisational structures wrapped around it. While generative AI and agentic models continue to evolve, the real bottleneck is enterprise readiness, or more accurately, the lack of it.
We’re in a moment where the pace of innovation is outstripping the ability for organisations to adopt it. If this gap isn’t addressed, AI agents risk becoming just another experimental tool confined to pilot projects and innovation labs. To unlock the next phase of value, leaders must confront the cultural, operational and architectural barriers standing in the way.
Innovation is advancing, mindsets are not
Enterprises don’t lack ideas or technical ambition. Many have pockets of innovation scattered across teams, often in the form of hackathons, pilots and proof-of-concept trials. But moving from experimentation to production is a different game entirely.
In too many companies, the mindset around AI is still cautious, experimental and siloed. Risk aversion, especially around data use and model reliability, prevents AI projects from scaling beyond the boundaries of innovation teams. At the same time, operational workers often lack the clarity or ownership to move from pilot success to embedded workflows.
The gap between experimentation and execution in agentic AI is widening. While 47% of UK businesses are piloting AI agents and recognise their potential to drive competitive advantage, trust in these technologies is eroding as companies move from testing to implementation. Confidence in autonomous agents has dropped to 27%, down from 43% the previous year. This growing disconnect between innovation teams and those accountable for business results makes it increasingly difficult to build momentum, scale successful use cases and demonstrate ROI in ways leadership can trust.
From gatekeeping to enabling
This cultural gap can’t be fixed by technology alone. It requires a different approach to leadership, one that shifts from gatekeeping innovation to actively enabling it across the business.
Leaders need to foster a culture that is not only open to experimentation but structured for action. That means creating psychological safety for teams to test and fail fast, while also setting clear expectations for how AI investments translate to business impact.
Equally important is building trust in the technology. Many frontline teams still hesitate to adopt AI agents because they don’t fully understand how decisions are being made. Transparency, auditability and alignment with governance frameworks must become foundational. If people don’t trust what an agent is doing or why it’s doing it, they won’t delegate decision-making and the entire promise of agentic AI collapses.
Agentic AI requires real-time infrastructure
Underneath the cultural and operational challenges lies a technical one: many organisations are not yet equipped with the infrastructure needed to support agentic AI.
Because large language models (LLMs) are stateless, agents require a reliable way to store and retrieve memory such as task history, user preferences or real-time context. They also need low-latency access to data and the ability to act across systems. Without this, performance suffers, and the risk of errors or hallucinations increases.
To scale agents beyond isolated pilots, enterprises need systems that support real-time decision-making and coordination across workflows. This often means modernising how data is delivered and accessed within applications, not just which models are being used.
The leap from experiment to enterprise-grade deployment is not automatic. It requires investment in the operating model spanning leadership, culture, architecture and data readiness. High-performance data storage systems, with sub-millisecond latency and massive scalability, provide the foundation needed for AI systems to retrieve and process memory at human-conversation speeds.
Some organisations are further along, even running agentic workloads in complex on-prem environments. But for many, the ability to scale will come down to collaboration across engineering, product and business teams, and a willingness to confront what is getting in the way.
Being future-ready
The question for enterprises is no longer ‘Can we build this?, it’s “Are we ready to use it effectively?” In many cases, the answer isn’t there yet.
But with the right cultural shift, leadership mindset and operational infrastructure, agentic AI can move from isolated pilots to a central driver of business value. The technology is ready; the enterprise needs to catch up.
