If you’ve been following the AI hype, you know that lots of companies now say they’re using AI in multiple areas. But the reality is that most of them are still in their early adoption days, far from full-range enterprise-grade scaling of AI tools.
According to McKinsey’s recent survey, which polled nearly 2,000 respondents across the globe, 88% of organizations report using AI in at least one business function. That’s significantly up from 78% a year ago, considering that the share of those non-adopters is not that large already.
However, when you dig deeper, you see the imprssive picture under a different angle: only about a third of companies say they have scaled their AI programs enterprise-wide, while most are still in the “experimenting” or “piloting” stages.
Where things get even more interesting is when it comes to AI agents everyone is so excited about these days. If you follow the media headlines, it seems that everyone has already deployed these innovative systems in their operations. The survey puts things into perspective:
-
truly, 62% of respondents say their organisation is at least experimenting with AI agents.
-
but only 23% say they are scaling at least one agentic system somewhere in their enterprise.
-
furthermore, even among those scaling, most are doing so in just one or two functions and in no single function do more than 10% of respondents report widespread agent deployment.
Why AI agent scaling might be not that easy
You might wonder: “If AI works, why don’t companies just roll it out everywhere?” The McKinsey data gives some hints.
Mixed results on bottom-line impact
-
While many report cost savings or revenue boosts on particular use cases, only 39% of respondents say AI has made any impact on enterprise-level EBIT.
-
And for most of those, the impact is small: less than 5% of total EBIT.
-
Yet, despite limited financial impact, 64% say AI is helping with innovation. Many also cite improved customer satisfaction and better competitive differentiation.
That means AI often delivers great operational value. But that value doesn’t always translate into big company-wide transformation or overall profitability. At least not yet.
Few firms treat AI as a transformation, not just a fancy tool
The firms that are seeing real enterprise-level returns from AI (call them “AI high-performers”) behave differently. These companies don’t just ask how this can save costs in a particular segment like IT. Instead, they aim for general growth and innovation, redesign workflows, and integrate AI into multiple functions, like marketing/sales, corporate finance, product/service development, etc.
They don’t use AI tools blindly, as well. The company’s management and expert professionals define when humans need to validate AI outputs because models can err or hallucinate.
That makes perfect sense, if you think about AI evolving from your search/chat buddy into a large-scale orchestrated system that becomes the backbone of your operations. In the latter scenario, it cannot scale with a siloed or careless approach, especially when it comes to high corporate responsibility areas such as fintech which handles miriads of sensitive data.
Agentic AI not only facilitates processes but also adds complexity
AI agents promise more autonomy and multistep workflows, which is undoubtedly exciting. But it also ups the stakes on reliability, process integration, and oversight. McKinsey’s findings show adoption of agents is highest in IT and knowledge management (e.g. research, data synthesis, and service-desk automation).
Those environments are treated as safe-ish to test agents. But putting agents into more critical or customer-facing domains means you first and foremost need trust, data hygiene, clear workflows, human-in-the-loop design, and maybe even rethinking what roles humans play.
Achieving that is not always easy. The technology is novel. There are no clear guidelines and rulebooks that show exactly what to do to get perfect balance between automation and trustworthiness. Without a clear roadmap, many organizations are raising risk mitigation efforts. More companies are now thinking about privacy, explainability, organizational reputation, regulatory compliance, and more besides basic convenience and cost savings brought by AI tools. Today, it’s basically an uncharted territory one should figure out how to navigate. Not every business has necessary time, skills or financing to do that. Thus, many firms just wait for the ready-made plug-and-play solutions similar to white-label software sets.
What “scaling AI agent strategies wisely” should mean: honest plan
As with every new solution or process introduced into the company’s workflow, before you roll out any agentic tool, ask yourself a few important questions:
-
What strategic outcome do we expect? Is it cost reduction, faster product development, better customer service, new product ideas, data-driven decision making or something else?
-
Which business functions have the right mix of repeatable tasks and potential upside for automation or augmentation?
- Do we have the right tools and skills for AI to bring enterprise-level value or are we just starting small, expecting lesser but still significant results?
One of the biggest findings from McKinsey is that successful scaling is strongly correlated with revising workflows altogether, not just layering AI on top of old processes, which for many businesses is just like patching a large wound without treating it first.
Revising workflows for AI deployment
Before you deploy an agentic solution or system, map out clearly how work will change, who supervises or validates results, and how you measure success. This is especially critical for functions like customer service, compliance, finance, etc., where mistakes or hallucinations have real impact calculated in greenbacks.
To truly scale the AI agent use across the whole enterprise operations, the firm’s senior leaders must be genuinely committed to, and own, corporate AI strategies.
To scale effectively, AI initiatives need:
-
leadership support and commitment,
-
allocation of resources (not just money but also data infrastructure, talent, governance, and so on),
-
cross-functional collaboration,
-
and strategic alignment (i.e. AI used where it fits, not where it’s convenient).
Without that, you risk of ending up with a scattered patchwork of pilots and little enterprise-wide impact.
Finally, wise AI scaling is not rushing to pick up every shiny brand-new tool. It’s largely about responsible deployment too.
Hence, when you’re giving AI tools the power to execute multi-step workflows, make decisions, or interact with customers (and I believe it’s rather “when” then “if”), you must first clearly decide when and how human-in-the-loop supervision happens, what KPIs or guardrails to monitor, how to handle errors, potential bias, ethical dilemmas or so-called “hallucinations,” and how to ensure transparency, accountability, and compliance.
Why now might actually be a good time to experiment with AI agents but not expect quick overhaul
The thing is, 2025 seems to be the moment when the dust of hype is settling a bit (at least, we do not hear much about metaverses anymore), and the real work of building useful, reliable AI systems is starting. We’re exploring the immense potential of both generative AI and coordinated agent systems, but don’t really expect AI to solve all the problems. That’s a good old pragmatic way towards realistic adoption with measured benefits and cautious optimism.
If your approach is balanced and thoughtful, with clear goals, solid processes, staged scaling, and proper governance, AI agents could indeed become powerful tools. But you’ve got to be in it for more than just quick wins. The companies that scale AI enterprise-wide are the ones thinking long-term, aligning AI with strategic priorities, and investing in the fundamentals.
