In a landscape buzzing with talk of an “agentic workforce,” it can be difficult to separate the hype from the reality on the ground. To cut through the noise, we sat down with Chloe Maraina, a Business Intelligence expert whose work lives at the intersection of big data and compelling business strategy. With her deep understanding of data science and a clear-eyed vision for the future, Chloe helps us understand the true state of AI agent adoption. We’ll explore the stubborn interoperability issues slowing down progress, what the “fragility” of today’s agents looks like in practice, and a concrete roadmap for IT leaders looking to move from small-scale pilots to meaningful implementation. Finally, we’ll look ahead to how this technology will reshape job roles for everyone, not just the tech-savvy.
The article cites a McKinsey survey where only 23% of organizations are scaling agents. Given the technical challenges and vendor “walled gardens” you’ve mentioned, what specific interoperability issues are preventing this wider, multi-function adoption? Could you share a common example you’ve observed?
It’s a source of immense frustration for so many teams. You have these incredibly powerful tools, but they’re designed to exist in their own little sandboxes. The core issue is that vendors are, understandably, trying to build and protect their own ecosystems and figure out how to monetize the data their agents generate. So, they create closed systems. A classic example I see is in retail. A company might have a fantastic customer service agent from one vendor that handles inquiries, but its e-commerce platform, from a different vendor, has its own agent for managing inventory and sales. The dream is for these agents to work together seamlessly, but their APIs simply aren’t compatible. The customer service agent can’t directly check real-time stock or process a return without a clunky, custom-built integration. This is why we see that low 23% figure for scaling; companies can get one agent working well in one department, but creating a truly interconnected, multi-agent system across the business remains a massive technical and competitive hurdle.
One CTO warns that agentic imprecision can create “unreliable junk.” In a software development context, what does this fragility look like day-to-day, and how do the agents’ current memory limitations contribute to these failures? Could you walk me through a typical scenario?
That “unreliable junk” quote really hits home for developers. Picture this: a developer is tasked with a complex coding assignment. They use an AI agent to help automate parts of it. In the short term, it works beautifully, generating clean code for a specific function. But the next day, when they ask the agent to build an adjacent function, it has no memory of the previous work. It’s like having a conversation with someone who forgets everything you said five minutes ago. Because the agent lacks that long-term memory and contextual understanding of the entire project, it might generate new code that inadvertently conflicts with what it wrote yesterday. This creates a subtle but significant bug. The developer then spends hours, sometimes days, hunting for this tiny point of failure. The initial time saved is completely lost, replaced by the deep frustration of debugging a mess created by the very tool that was supposed to help. That’s the fragility in a nutshell; even a tiny bit of imprecision can derail the entire process.
A CIO you spoke with is confident that agent implementations can grow from 2% to 20% as the technology matures. For an IT leader starting a pilot today, what are the first three practical steps for building the necessary core abstraction layer and governance framework before attempting to scale?
For any IT leader dreaming of that jump from 2% to 20%, it’s all about laying the right foundation. You can’t just let a hundred agents run wild and hope for the best. The first, and most critical, step is to build a core abstraction layer. Think of this as a universal translator for your agents. It ensures that you aren’t permanently locked into a single vendor’s ecosystem and that different agents can, at a basic level, communicate. Second, you must establish basic orchestration. This is about defining the rules of engagement—how agents are triggered, what their priorities are, and how they hand off tasks. It’s like being the conductor of an orchestra; you need to make sure everyone knows when to play their part. Finally, and this is non-negotiable before you scale, you have to build in robust governance and monitoring capabilities from day one. You absolutely need visibility into what these agents are doing, the data they’re accessing, and their performance. Without that control, you’re flying blind and inviting risk.
IDC forecasts that by 2026, 40% of Global 2000 job roles will involve working with AI agents. What will this “redefined workstream” actually look like in a non-technical department like marketing or finance, and what key skills should employees start developing now to prepare?
That 40% figure sounds dramatic, but it’s more of a shift in focus than a replacement. In finance, for example, an analyst won’t spend the first week of the quarter manually pulling data from ten different systems. Instead, they will direct a team of agents to gather, clean, and structure that data into a preliminary report. The analyst’s job then becomes about higher-level thinking: interrogating the data, spotting the anomalies the agent flagged, and building the strategic narrative around the numbers. In marketing, a campaign manager might oversee an agent that runs hundreds of A/B tests on ad copy simultaneously, optimizing for performance in real time. The manager’s skill is no longer in the manual setup but in defining the strategic goals, setting the creative boundaries, and interpreting the results to plan the next big campaign. The key skills to develop now are strategic oversight, critical thinking, and the ability to ask the right questions. It’s about learning to be a manager of digital workers, not just a doer of manual tasks.
What is your forecast for agentic AI?
My forecast is one of cautious, and often messy, optimism. The idea that by 2026, entire businesses will be running autonomously on agents is pure hyperbole. That idyllic dream is still a long way off. However, we are going to see a significant push beyond the current 39% of companies just experimenting. The reality will be mixed. We’ll see incredible, tangible successes where agents are scaled within a single business function, like IT support or customer service, to automate entire workflows. But the ambition to create multi-agent systems that work seamlessly across different vendor platforms will continue to be a major struggle. The collective business imperative will absolutely drive the industry to solve these challenges, but for the next few years, the story won’t be about a mainstream takeover. It will be about redefining specific workstreams and learning, often through high-profile failures, how to build the technical and governance foundations for a truly agentic future.
