What enterprises need to make Agentic AI successful
Recently I’ve been asked by several executives what it takes to make Agentic AI truly successful in an enterprise. By Agentic AI we mean solutions that consist of specialist AI agents that are orchestrated together to accomplish higher level tasks such as undertaking end-to-end business processes. Most readers will be familiar with the analyst forecasts – for example Gartner expects $450 billion of revenue to be generated for firms through Agentic AI by 2035. In my experience though, the real story lies in what’s happening inside organisations today. I’m seeing a surge of interest from CXOs, driven by a paradoxical truth: many stand-alone AI initiatives have failed to deliver the ROI executives expected.
Some of the research (e.g. MIT's Project Nanda) is disputed, but there is a direction of travel that we have seen ourselves. That track record of disappointing results is also what makes Agentic AI so commercially compelling. The very issues that caused standalone AI solutions to fall short – such as cultural barriers to adoption, and disconnected systems – are exactly those that agentic AI, when designed and implemented correctly, is designed to solve.
Driving adoption through culture and design
One of the toughest barriers in AI projects is business adoption. You might add an AI tool into an existing process, but unless the workforce chooses to use it, you won’t see any value added. With agentic AI, we flip the model: employees become orchestrators of the agents, not just passive users. Adoption is embedded by design.
That said, cultural change takes time; people need to embrace AI as a colleague, not a threat to their skills. Similarly, you can’t just drop agents into a broken process and expect magic – you need to first transform the process and the mindset. At NTT DATA, we focus on end-to-end designs that encompass people, process and technology to help organisations build the full transformation story.
From islands of innovation to an integrated engine
Another major challenge lies in the tangle of disconnected systems many organisations are now stuck with. In the past, too often the proliferation of isolated AI tools has turned people into the glue that’s holding everything together - a role which is neither scalable nor efficient. This time, we have an opportunity to do things differently. To unlock the full potential of agentic AI, technology must be interoperable by design, allowing agents to share information and build on each other’s outputs.
We describe this as the flywheel effect; in this scenario, the the more agents you create, the more value you should be able to generate. But success here is not a foregone conclusion: even with agentic AI, it’s possible to create ‘agent sprawl’ – introducing multiple, disconnected technologies that force humans back into the role of integrator, instead of letting an interoperable backbone do the work. The warning signs are already there, fuelled by executive pressure to “move fast” and development teams’ fear of missing out. Without alignment, agent sprawl risks slowing the very progress it aims to accelerate.
Skillsets matter but so do mindsets
A third challenge lies in developing both the technical skills and the mindset in engineering teams to make agentic AI work. New toolkits are lowering the barrier to entry, from drag-and-drop environments to low-code platforms. We’re collaborating with partners such as OutSystems to explore what these tools can unlock, while every major hyperscaler is racing to build its own agentic capabilities.
However, real progress happens when teams learn how to think differently. That means understanding how to use augmented engineering tools to design systems where humans and agents build, iterate, and learn together. That mindset shift will separate organisations that experiment with agentic AI from those that truly transform through it.
Where humans and agents co-create
Despite the challenges, I’m already seeing clients turn the promise of agentic AI into something tangible. In one project, for instance, we’re using agents to take on complex knowledge work, such as designing new software architectures. Each agent tackles a different piece of the puzzle, then reviews and refines the others’ contributions in continuous cycles of improvement.
The human in the loop steps in at the end as a collaborator who shapes the final design and, in doing so, teaches the system to evolve. Every interaction becomes part of a feedback loop, creating an ecosystem that grows smarter, more creative, and more capable with each iteration.
What we’re now seeing – something that’s been missing previously – is a type of co-creation: a glimpse of an enterprise in which human ingenuity and agentic intelligence feed off one another, sparking new ideas and accelerating progress.
If you’d like to know more, take a moment to get in touch and arrange a 45-minute consultation.