As AI adoption surges across industries, enterprise leaders are racing to weave the technology into every layer of their software stacks.
But while enthusiasm at the executive level is high, engineers at the frontlines are warning of legacy system constraints and data blockers once again throttling transformation.
For many data engineering teams, the daily reality isn’t about building next-generation AI models, it’s about making them fit within ageing, rigid systems, and this isn’t just a passing frustration.
According to AND Digital’s Know Me or Lose Me report, 56 per cent of business leaders are planning to invest in AI despite knowing their data may not be accurate, with 77 per cent of senior engineers reporting that integrating AI tools into existing applications presents a significant pain point.
Chief for Data at AND Digital.
The AI gold rush is exposing deep structural issues in enterprise tech, from legacy systems and data chaos to a widening skills gap.
Furthermore, given the commoditization of AI applications, competitive advantage such as truly personalized customer experience is more demanding if the need for integration to richer data sets is held deep within an organization. Leveraging data is non-negotiable, yet legacy lock in means it’s elusive.
Legacy systems with modern pressures
At the heart of the data surfacing challenge lies an uncomfortable fact; many companies are still heavily reliant on legacy systems.
These are the systems that keep the lights on by powering supply chains, managing customer records, or handling financial transactions. But they were built long before today’s AI tools were conceived, and weren’t designed to interface with them.
These older systems often operate on outdated architectures and siloed data, making AI implementation not only expensive but fragile and making legacy dependencies strategic liabilities.
Ageing IT infrastructure is not only delaying digital transformation initiatives, but actively holding back broader AI strategies. In a market where first-mover advantage can be critical, falling behind in AI readiness could mean losing competitive edge entirely.
The global market for AI application development is expanding rapidly, valued at $5.2 billion and projected to grow substantially. As a result of this, the space has become a playground for startups and major cloud providers all promising to simplify AI deployment.
But while platforms that reduce AI integration are in high demand, they are not silver bullets. Choosing the wrong tools, or adopting them without the correct framework can amplify existing problems.
The human factor in AI success
There’s a growing disconnect between how business leaders view AI and what it takes to actually implement it.
From the top, AI often looks like an opportunity for transformation, faster processes and smarter decisions, but for the engineers who are in charge of delivering those outcomes, the focus is on feasibility, ethics and infrastructure. Too often, the push for rapid rollout comes without an equivalent investment in skills or support.
Engineers and data teams aren’t just plugging models into apps, they’re navigating complex situations around data privacy, model accuracy and ongoing maintenance. These tasks require both technical fluency and organizational alignment, but few companies have invested enough in bridging this gap.
Many organizations prioritize the fast deployment of AI but overlook the readiness of their workforce and the quality of their data. You can have the best tools in the world, but if your teams don’t understand how to use them or don’t trust the data, your AI won’t deliver lasting value.
This is why upskilling remains one of the most critical and under-addressed challenges in AI integration. It’s not enough to have a handful of machine learning specialists, organizations need developers who understand how AI affects software and can work with evolving models.
The companies that succeed with AI in the long run will be those that recognize this early, not just investing in tools, but empowering the people who use them.
Data readiness is non-negotiable
Underpinning all of this is one essential truth; no AI system can outperform the quality of the data it’s built on. And yet, data remains one of the weakest links in most organizations’ AI strategies.
Inconsistent, siloed, or outdated data is more than an inconvenience, it’s a direct threat to model reliability, system integration, and ultimately, user trust. AI that’s trained or deployed on poor data doesn’t just underperform, it can actively mislead, making flawed predictions or reinforcing bias.
Still, many companies continue to pour resources into AI initiatives without first addressing their data problems. They focus on what AI might do, rather than whether their infrastructure is prepared to support it.
Clean, well-structured and accessible data unlocks the true potential of AI and also reduces the burden on engineers to make integration smoother, more predictable and more scalable.
For organizations truly serious about AI this must be the starting point to understand where data lives, how it flows, who controls it and how it can be trusted. High-performing AI needs high-performing data.
The AI revolution is real and the stakes are high, but achieving meaningful results requires more than ambition, it requires discipline and the right tools. Integrating AI into enterprise environments isn’t about quick wins, rather it’s about modernizing infrastructure and building clean, reliable data foundations.
As companies look to scale AI efforts, they must listen carefully to the engineers and technologists doing the work behind the scenes. Their warnings are not resistance, they’re insights, as without the right groundwork even the most promising AI projects can fail.
But ultimately, success won’t come from tools alone. It comes from those who take the time to build well and build smart.
We list the best COBOL online courses.
This article was produced as part of TechRadarPro’s Expert Insights channel where we feature the best and brightest minds in the technology industry today. The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here:
