Why 80% of organisations have yet to see results from GenAI and how Agentic AI can make the difference, if approached as a strategy rather than a gimmick.
AI is everywhere. In pilots, PowerPoints, and experiments. But who is actually seeing real results?
According to McKinsey, nearly 80% of organisations are now using generative AI, yet almost the same percentage report no significant impact on profitability. The phenomenon has even been given a name: the GenAI paradox.
So what’s going wrong? And more importantly: what does Agentic AI actually promise – and does it deliver?
What is Agentic AI (and why is it not just a buzzword)?
According to Eliseo Ferrante, Visiting Professor of Computer Science at New York University Abu Dhabi and Assistant Professor at the Vrije Universiteit Amsterdam, Agentic AI is: “Agentic AI is a system in which one or more autonomous AI agents interact with each other and with other software tools.”
These agents can collaborate directly or through integrated systems, enabling them to perceive, reason, act, and learn individually or collectively within complex digital environments. Agentic AI carries out entire sequences of tasks that would normally fall to a human, including interpretation, coordination, and feedback.
So where does the real potential lie?
Agentic AI may sound like science fiction, but the technology is already here. The challenge? Making it work in practice.
One of the most compelling showcases comes from Uber. During LangChain’s Interrupt conference, Sourabh Shirhatti and Matas Rastenis from the Uber Developer Platform Team demonstrated how Agentic AI is radically transforming the productivity of 5,000 developers. They were looking to solve the problem developers enjoy the least: writing tests and reviewing code.
Their answer? An agentic tool that automatically detects errors and security issues and proposes solutions. The system performs thousands of interactions daily and has already saved 21,000 developer hours.
Not a gimmick. But because it actually works.
Why isn’t this working (almost) anywhere else?
Simply put: because we’re using it the wrong way.
McKinsey notes that many companies are still applying AI to make existing processes slightly smarter – instead of redesigning those processes around what agents can truly do. As a result, Agentic AI ends up being a fast assistant in a slow system.
Automation consultant Jelle Acda puts it clearly: “There are still hardly any real Agentic AI use cases in practice. There’s plenty of intelligent automation, but not true agents.”
And where such systems do exist, decision-makers often don’t know where to begin. Legacy infrastructure, lack of governance, and limited organisational readiness are just some of the reasons why C-level leaders still say: “Not just yet.”
When does it actually work?
Some organisations don’t see Agentic AI as a gimmic, they see it as a chance to fundamentally change how they operate.
At DELTA Fiber, under the leadership of Wendy Persoon, Digital Strategy Lead, two proof-of-concepts are currently underway in which Agentic flows are being tested. One application automatically applies the correct discount for customers during and after their contact with the customer service centre—a strong example of Agentic AI, where traditional RPA proved insufficient.
A second application responds to and processes queries about invoices more quickly and accurately than the standard chatbot, which is based on predefined question-and-answer logic. Although it is still too early to speak of a full-scale rollout, Wendy notes that limitations which posed obstacles six months ago have now largely been resolved.
DELTA Fiber is actively exploring whether this technology can be applied more broadly across other business processes.
An impressive real-world example at scale comes from Similar.ai, where we spoke with co-founder & CEO Robin Allenson. His software helps large e-commerce platforms create the best experience for the most users with the fewest webpages possible. It automatically generates new product category pages, takes care of internal linking and cleans up and replaces underperforming pages for new ones. All based on real-time search and demand data.
This used to be manual work, taking hours per page. Now, thanks to an agentic architecture, Similar.ai can generate thousands of pages for a web shop with hundreds of thousands of products, fully autonomous, at speed. For those who don’t yet feel fully comfortable with an autonomous system, it can also be set up with human oversight.
And that’s not all: the same agents continuously monitor whether internal links remain relevant and up to date. Outdated links are detected and replaced with a more contextually appropriate alternative.
Although Robin rarely markets his product as “Agentic AI”. “The customer just wants something that works” it is, at its core, precisely that: autonomous, proactive, domain-specific, fast and scalable.
What do these examples have in common?
✅ They start with the problem, not the technology.
✅ They’re functionally embedded in the process.
✅ They’re driven by people who understand what it takes to make it land.
Finally: Stop building blindly. Start building with insight.
At NXTminds, we believe in technology that serves people. Not in buzzwords, but in practical solutions. That’s why we connect organisations with experts who think beyond tools – and help take the right steps towards scalable AI impact.
👉 Curious what that looks like in your organisation? Let’s talk.
References: McKinsey: Seizing the Agentic AI Advantage, LangChain Interrupt: How Uber Built AI Agents That Save 21,000 Developer Hours with LangGraph