Enterprises have spent the past few years experimenting with AI and agentic applications, but the bigger challenge is in turning those experiments into production systems.
Speaking to Tech Arabia at LEAP 2026, Sid Bhatia, Area VP & General Manager, Middle East, Turkey & Africa and India, Dataiku, said organisations need to shift their focus from simply building AI applications to putting the right business, governance and operational foundations around them.
“Experimentation is good, but the real question is how many organisations are actually able to take this to production? How do you convert that experimentation into measurable business impact?” Bhatia said.
He said the starting point should be identifying what the organisation wants AI to achieve, rather than beginning with the technology itself. “It could be around reducing costs, increasing revenues, launching new products or services, or more recently, AI monetisation.”
Bhatia explained that once firms have identified those value drivers, it is important to come up with the top use cases that they can quickly implement and demonstrate value against.
He also cautioned against treating the number of AI agents an organisation builds as a measure of progress. “We need to move away from this race of creating 500 or 1,000 AI agents. The companies that will be successful are the ones that have a production-ready mindset,” he argued.
That means treating governance, security, orchestration, model monitoring and maintenance as part of the deployment process rather than adding them later. Bhatia also said business teams need to be involved alongside data scientists, data analysts and data engineers. Once an AI use case delivers results, organisations should be able to capture and communicate that value internally.
Governance moves to centre
As AI adoption grows, Bhatia said governance has become one of the biggest concerns for enterprises. While building AI agents is becoming easier, organisations need to consider what happens after those systems are deployed.
“Six, 12 or 18 months down the line, would you be in a position where you can confidently say that you control AI, or does AI control you, or is it real AI chaos?” he questioned.
‘We need to move away from this race of creating 500 or 1,000 AI agents. The companies that will be successful are the ones that have a production-ready mindset’
For Bhatia, an enterprise AI agent needs to go beyond simply performing a task. It needs to deliver measurable business value while also being trusted by the organisation.
“The right kind of governance has to be there for the AI agent to be successful and adopted across the organisation.”
Managing models, costs and data
Dataiku is also focusing on the challenges that emerge as organisations build applications on large language models (LLMs). Bhatia pointed to Dataiku’s LLM Mesh, which allows organisations to move between different LLM providers rather than having applications tightly tied to a single model provider.
“Dataiku LLM Mesh helps you seamlessly move from one LLM provider to another. That creates flexibility and, therefore, a lot of trust,” he said.
Cost is another consideration. Enterprises need to account for API and tokenisation costs and understand how those costs could change before putting an AI application into production. “You need to give customers the ability to do what-if analysis. If I put this agent into production and compare it across various LLM providers, what is the cost going to be?”, he warned.
Data security and compliance are also part of that equation. Bhatia said organisations need controls around confidential and personally identifiable information to avoid exposing sensitive data outside the organisation.
“This is where the LLM Mesh really comes in from a governance angle. Governance and control create trust, and trust is what the agents need. That is how you productionise,” he added.
Turning enterprise knowledge into AI
Beyond governance, Bhatia sees an opportunity for organisations to bring their existing business knowledge into AI systems.
Enterprises have accumulated years of operational knowledge and expertise, but the challenge is converting that knowledge into AI workloads that can work with existing applications.
“How do you convert the data, expertise and tribal knowledge that you have built over 30, 40 or 50 years into an agentic AI workload that can be orchestrated and seamlessly integrated with your business applications to result in measurable value?” he said.
He explained that AI sovereignty is now becoming an important regional conversation, alongside governance and the need for enterprises to retain control over how AI is deployed.
For Dataiku, the focus remains on bringing together people, orchestration and governance as organisations move from experimentation to wider AI deployment.
“The whole idea of having the right people work together in a collaborative manner, while making sure that you have the right orchestration and, of course, the right governance, is what differentiates organisations that are able to move ahead with AI,” Bhatia said.