Senior IT leaders from a variety of industries, ranging from publishing to wealth management and banking, gathered for an executive roundtable dinner in New York City to discuss the use of AI in application modernisation and quality engineering. The event took place in early May, and was jointly hosted by GlobalData and Hexaware, a US based IT services company.
GenAI and agentic AI took centre stage at a roundtable event in New York as attendees discussed how technology will move their organisations towards greater automation. While acknowledging the numerous benefits of AI, executives were nonetheless quick to acknowledge that the path forward won’t be easy to navigate and includes its fair share of challenges.
All participants agreed that AI technology is evolving at a rapid pace, moving from predictive AI, to GenAI, and on to agentic AI at breakneck speed. GlobalData predicts that the global AI market will grow from $182bn in 2025 to $888bn in 2029. Consulting services drive a substantial portion of overall market growth.
Organisations that were reluctant to work with a third party in the past are expected to embrace the professional services offered by a range of suppliers in order to better manage the complexities associated with agentic AI and multi-agent systems. Additionally, AI is an equal opportunity technology; it will be embraced by almost all vertical industries.
Identifying ROI is a key challenge
The most significant hurdles when adopting AI include calculating the return on investment (ROI), addressing security concerns, and managing the AI skills gap. In prior conversations some IT leaders have revealed that they are hesitant to dive in head-first with AI and would prefer to wait for the dust to settle before tackling advanced applications of technology.
Although this approach is understandable, waiting is a mistake no one can afford to make. At the executive roundtable in New York City, participants encouraged each other to push forward, noting that experimentation is critical to establishing a foundational knowledge base and crafting an AI strategy.
Those that delay or are overly cautious don’t do themselves any favours. They will miss key steps in the learning process and be left behind. Furthermore, board members, C-level executives, and lines of business are pressuring IT teams to step up the pace of AI deployments. All roundtable attendees agreed that balancing the demands for more AI-driven workloads by stakeholders with the complexities of deploying the technology responsibly and at scale is a massive challenge for IT decision-makers.
Developer use of AI yields robust ROI
AI ROI is a hot button topic in the tech world today. Numerous highly publicised surveys have yielded widely varying results on the measurable impact of AI projects. Not surprisingly, the subject also sparked a lively conversation during the roundtable discussion.
Many asked if they should change their approach to ROI analysis or if they are measuring the correct metrics; all were eager for guidance in developing ROI frameworks for their projects. Findings from a recent GlobalData poll on ROI, by use case, noted that AI adoption by developers is one of the most successful applications of AI, with almost half of respondents anticipating a positive ROI within one year.
Among roundtable participants, the cost of AI processing was also a topic of interest, with organisations noting the complexity of predicting inference costs. Token economics is becoming a greater concern as companies enter a world in which the use of AI is increasingly exponentially, yet the mechanisms to manage inference costs are lagging.
AI for app modernisation extends beyond coding
Participants noted that they have seen great success with using LLMs to generate, refactor, or optimise code. By reducing manual effort and cycle time, teams can modernise faster with fewer hours spent on routine tasks—driving down overall cost. AI can provide support for legacy code, especially where developers with the necessary skillsets are scarce.
But the use of AI by developers isn’t limited to coding; the technology can support multiple phases of the application modernization process and support quality engineering. Agentic AI can explore systems, correlate data, and explain what components do and why they behave the way they do. It helps uncover data access patterns, dependencies, and integration points, and can flag inconsistencies—reducing risk before making changes. When necessary, AI agents can create documentation of processes
AI can also be used to automate testing and support quality assurance. AI can automatically generate regression tests to make sure new code doesn’t harm existing functionality, accelerating the quality assurance process and reducing defects.
It also supports incremental modernisation, allowing for systems to remain in production while updates are made, minimising downtime. With faster discovery, automated refactoring, and stronger automated testing, modernisation can shift from being a one-time project to a continuous, incremental initiative. Furthermore, engineers spend less time on repetitive tasks and more time on more strategic, higher-value, projects.
Observability is critical to responsible AI
Attendees at the roundtable event also voiced frustration over the reluctance of some employees to embrace AI. The need to build trust in AI solutions is critical to assuaging user concerns and promoting greater adoption. Furthermore, responsible AI underpins trust and is essential for adoption at scale. Alarmingly, in a recent GlobalData poll only 30% of respondents were confident that their company had a responsible AI strategy in place. There are many elements that contribute to responsible AI, including data management and governance, regulatory compliance, security, end user training, and model observability and management.
While many of these elements have been around for years and applied when working with predictive AI, (explainability, data quality, bias concerns) they are now being expanded to encompass gen AI and agentic AI. Responsible AI now includes the need for observability that monitors agents, the data they access, the other agents they communicate with, who is accessing them, and the actions taken by agents. Observability tools need to incorporate appropriate escalation triggers, rules for exceptions, and guardrails, as well as ensure AI is performing as intended.
Enterprises are all in the same boat
Events such as the NYC roundtable provide IT leaders the opportunity to share experiences and provide assurance that peers are grappling with similar challenges. Practitioners across industries are struggling, yet they are moving forward despite some discomfort. Almost all organisations are short on skills and trying to learn from peers, even peers from outside of their industry. And finally, they know they need to rethink processes to best leverage human vs AI’s strengths, and not just layer agentic AI onto existing workflows. But, the majority are only in the early stages of this mindset shift.
