By Contributor,Gary Drenik

Copyright forbes

Artificial Intelligence
Teera Konaka – Getty Images

Sometimes, it’s important to step back and look at the wider picture. Amid the enthusiasm for the potential of new technologies, and despite years of digital investment, productivity is stagnant or decreasing. Automation has helped teams move faster in some areas, but not necessarily more effectively. GenAI pilots are so far delivering targeted impact, but many are not advancing to production, remaining in pilot wonderland and of little benefit to their sponsors.

A slow-emerging but unstoppable truth is tempering boosterism: to convert the potential of AI, machine learning, and automation into real, positive business outcomes, organizations need more than incremental task efficiency; they need systemic reinvention. To state it another way, businesses will have little or no benefits from these technologies without reinvention and business-led changes to their processes, people, and structures.

From tasks to journeys

Technology has long excelled at automating repetitive tasks, such as data entry or password resets. More recently, new architectures have enabled the understanding and generation of natural language to help with tasks like drafting emails or executing multi-step flows. Agentic AI goes further, or aims to—powering autonomous agents that can plan, act, observe, and learn across entire workflows and customer journeys.

What has yet to land for many is the distinction between the two. Automation is a lever, while Agentic AI is a new operating model. Intelligent agents aren’t just smarter bots; they act as composite, reusable work resources that can be orchestrated across various domains. In this sense, the impact doesn’t come from agents or from models and architectures alone, but from the way they reason through problems and their access to internal and external systems. Underlying protocols like MCP and A2A are quietly powering this shift – often invisible, but foundational.

The shift is hard to scale

The high costs of infrastructure, integration, and the talent necessary to make the leap are some of the reasons that, while most companies today are deploying agentic workflows of some kind, few have deployed true autonomous AI agents. And this despite individual adoption of GenAI being at markedly higher levels. According to a recent Prosper Insights & Analytics survey, over a third of over 18-year-olds in the US are already using GenAI, in addition to some 44 percent of Business Owners, 50 percent of Executives, and 31 percent of Employees.

MORE FOR YOU

Prosper – Heard of Generative AI
Prosper Insights & Analytics

Not only do corporate employees use GenAI at lower rates than teenagers, but companies that have piloted GenAI projects are seeing pilots failing at an alarming rate, finds a recent study by MIT. Oana Cheta, a strategy and operations partner at McKinsey, acknowledges that some companies are struggling to implement GenAI, but holds that the source of their difficulties has little to do with the technology itself.

“The lack of successful implementation is not an AI problem, it’s an enterprise problem. AI isn’t overhyped. Enterprises are under-wired and under-equipped. In a few years, it will be invisible infrastructure. The winners won’t be the ones stuck in endless pilots but those bold enough to scrap outdated processes and fully integrate AI.”

Building blocks, not magic tools

It’s worth name-checking some of the usual (and less usual) suspects when it comes to low levels of successful AI progress. Organizations lack end-to-end journey redesign, especially at the early “blueprint” stage, which is composed of defining the current state, articulating the problems to solve, crafting the solutions, and (only) then redefining the business foundations. A lack of such rigor means efforts to encourage adoption are too often piecemeal, leading to isolated pools of experimentation with no overall coordination. There is also an overemphasis on model choice over fundamental architectural design, and investment is chronically underweight when it comes to the orchestration or memory layer aspects necessary to make meaningful progress.

Leaders who are building successfully in this area are treating AI as modular, reusable building blocks, not magic tools. Instead of chasing the latest model, they build reusable patterns: orchestration layers, agent libraries, decision protocols. McKinsey’s Oana Cheta points to instructive recent case studies.

“We accelerated the Agentic AI product blueprints for several global MedTech leaders. Modular agentic AI doesn’t just solve today’s service challenges. It gives MedTech leaders a scalable architecture for reinvention that can power growth, safeguard compliance, and set a new standard of consumer trust across the enterprise—all the while driving future deployments given the reusable nature of the code.”

“In a further engagement, we helped an AI-native, leading tech services platform to maximize its outreach to employers and drive growth in three phases—from strategy, through testing, to developing and deploying dynamic performance management and tracking. This, together with strong performance management and new ways of working for employees, has delivered measurable and significant uplift in revenue, increases in speed, and improvements in productivity.”

Dual-speed transformation

Strategically, and though it’s far from simple, focusing on both near-term wins and structural reinvention at the same time is how leaders and organizations can start to get more, and better, results.

Using atomic agents to power summarization, intent prediction, and content generation are good examples of where to get the momentum going. Embedding learning-by-doing into delivery teams is another instance where near-term initiatives can develop without significant capital investment. The inviolable law here is to avoid pilot sprawl by sticking to design principles. In fact, not doing this represents the biggest risk to transformation efforts, spelling almost certain failure.

At the same time, longer term, structural reinventions need to be addressed, involving the redesign of journeys, processes, workflows, and tasks when onboarding employees or customers, for example, or managing field service work. These are examples of the major rewiring that needs to happen at the core of businesses. And it’s hard. While it’s easier, for instance, to debate in detail the merits of a model or provider, it’s the reinvention of core processes that will eventually pay lasting dividends. Standing up small, multidisciplinary teams that can build and own agent-based solutions is one route to success. Aligning business and tech teams is another, as they tend to want different things.

So, what do we learn from taking a step back?

It’s time to deliver measurable, coordinated outcomes as part of a wider, structural change, not just demos. Agentic AI can and should serve as a blueprint for a better decision-making system and operating model. And those who can manage, wrangle, instigate, and breathe life into that better system will be the organizations that embody the future of work and business. As Oana Cheta says, “The companies that pick the right tools and rethink their operating model, not just their tech stack, will unlock productivity and tomorrow’s business and customer experiences at scale.”

Editorial StandardsReprints & Permissions