As enterprises accelerate their adoption of agentic AI, questions around governance, trust, data sovereignty and measurable business outcomes are becoming increasingly important. In an interview with Intelligent CIO Middle East, Babak, CTO of Cognizant, discusses the barriers preventing organisations from scaling AI, the growing challenge of agent sprawl and how Cognizant is deploying interconnected AI agents across its own operations.
Despite the growing excitement around agentic AI, many organisations are struggling to move from pilots to production. What do you see as the biggest barriers to enterprise-wide adoption?
We are seeing uneven adoption across organisations. The initial wave was largely focused on personal productivity, but there has also been significant uptake across the software development lifecycle.
However, when it comes to broader organisational augmentation, progress remains uneven. Some organisations are considerably further ahead than others.
Two of the biggest inhibitors right now are trust and cost. Enterprises need to be confident that these technologies can operate reliably within their environments while also delivering sufficient value to justify the investment.
With the emergence of shadow agents and agent sprawl, governance is becoming a major concern. How should enterprises approach the challenge of monitoring and managing AI agents?
I believe real-time governance and real-time guardrails are the way forward. However, there is a cost associated with implementing these capabilities, and effective governance needs to be centralised.
Large organisations will inevitably deploy agents from multiple vendors for different business purposes. All these agents will need to be registered and operate within some form of standardised or unified protocol.
The challenge is that we do not currently have an established standard for this.
Even before considering governance, organisations are already experiencing difficulties because they are introducing too many agents that cannot communicate with one another. This lack of interoperability is contributing to agent sprawl.
As adoption accelerates, enterprises will need a unified approach to managing and connecting agents while ensuring appropriate oversight.
Many organisations have invested heavily in AI but are still struggling to demonstrate measurable ROI. Where do you think they are going wrong?
In many cases, organisations have rushed to adopt AI without clearly defining their KPIs or expected return on investment. I think that is the wrong approach.
The starting point should always be the business outcome. Organisations need to establish what they want to achieve before introducing AI and then evaluate whether the technology is delivering the expected results.
That discipline has often been missing from early AI initiatives. However, we are beginning to see a shift as organisations become more focused on measurable outcomes and the business value of their investments.
How is Cognizant using AI internally to transform software development, application modernisation and enterprise operations?
Cognizant was the first, and probably remains the only, services company to publicly disclose the percentage of code it writes using AI.
We have achieved widespread adoption across the organisation. Last year, we organised a hackathon that broke a Guinness World Record, bringing together our teams and technology partners to educate employees on AI and familiarise them with different AI systems.
We have also been pioneers in identifying opportunities to transform our own internal processes.
Today, we are probably operating the world’s largest enterprise agentic system internally through our One Cognizant intranet.
The platform has more than 250 interconnected agents spanning HR, finance, IT, legal and sales. These agents are integrated across the organisation and are being used extensively by our 265,000 employees.
When you consider an environment of this scale, particularly one involving agents from multiple third-party providers, the need for effective governance becomes clear.
We developed our own trust framework to address these requirements and enable real-time governance across our internal agentic environment. We have now packaged that framework for our clients.
In many ways, we have been using our own technologies and solutions internally before taking them to market.
The enterprise AI market is becoming increasingly crowded. What differentiates Cognizant from other technology providers, particularly when it comes to delivering measurable business outcomes?
One of our biggest strengths is our deep understanding of our clients. In many cases, we know their businesses extremely well because we have been working with them for years.
We understand their operating environments, business processes, areas for improvement and the outcomes that matter most to them.
That positions us as a strong partner in their agentification journeys.
Many technology providers are bringing raw AI capabilities to the market. However, contextualising those technologies and engineering them into solutions that deliver meaningful business outcomes is a different challenge.
That is where Cognizant brings significant value. We understand how to connect technology with business context and translate it into measurable results.
We also have extensive experience in modernising enterprise systems, building unified API layers and integrating data across complex environments.
These are all foundational requirements for agentification, and they are capabilities we have been delivering for our clients for many years.
Much of the enterprise AI discussion has centred on GPUs and computing infrastructure, while the importance of data foundations has received comparatively less attention. How critical is the data layer to successful AI adoption?
The data layer is extremely important, but I also believe that improving data foundations and adopting AI agents can happen in parallel.
If an organisation’s processes are not functioning efficiently, even when performed manually, or if its data is fragmented across spreadsheets and individual systems, introducing AI will not automatically resolve those underlying problems.
Organisations need to address the challenges within their data environments because fragmented and poorly managed data will continue to constrain their operations.
However, that does not mean they should wait until their entire data infrastructure has been modernised before introducing AI agents.
These challenges already exist within their manual processes. AI agents can be introduced to augment those processes while the organisation simultaneously improves its data foundations.
The two initiatives should progress together rather than one becoming a prerequisite for the other.
Data sovereignty and sovereign AI have become important priorities across the Middle East. How prominently do these issues feature in your conversations with regional customers?
They certainly feature in our discussions.
I was recently looking at TAMM, the government services application used extensively in Abu Dhabi, and I was impressed by its capabilities and the feedback it receives.
Platforms like this already have multiple agents operating within their environments, with plans to introduce additional agents over time.
For organisations managing sensitive information and critical services, relying entirely on third-party agents operating in external environments is not necessarily an option.
There have also been concerns about frontier AI models anonymising user data and using it for training purposes. That raises important ethical questions about how enterprise information is handled.
For platforms such as TAMM, it is essential to ensure that both the AI capabilities and the underlying data remain sovereign.
Organisations need confidence that their information is protected and is not being transferred or exposed to external environments without appropriate controls.
Finally, what advice would you offer CIOs in the Middle East as they prepare their organisations for the next phase of agentic AI adoption?
I believe CIOs have an opportunity to take ownership of agentification across their organisations in a way that is both secure and empowering.
Their role should not simply be to control the adoption of AI, but to enable business units and teams to introduce agentic systems that can improve their operations.
At the same time, CIOs need to ensure that these systems are interconnected and can communicate effectively with one another.
The CIO, working alongside the CISO, must also take responsibility for establishing the governance, security and trust frameworks required to support these environments.
Ultimately, the opportunity is to empower the wider organisation to adopt agentic AI while ensuring that everything operates within a secure, interconnected and well-governed enterprise environment.
