'AI Enablement Requires Engineering'

'It requires tail layering and customising, and you can't just sprinkle your organisation with co-pilots and be done with it.'

There is a bubble in AI when it comes to building large language models (LLM) with big funding but the adoption of AI by enterprises will remain unaffected even if it bursts in the next two years, according to Cognizant's chief AI officer.

Computer scientist Babak Hodjat, widely known as the co-inventor of natural language technology, which contributed to the development of Apple's virtual assistant Siri, said there was a misconception in those who thought they were going to build super intelligence or artificial generative intelligence.

And whoever builds it first dominates the world.

"It's like sci-fi," he said in an interaction with Business Standard.

Concern about an AI bubble, and whether it resembles the dotcom bubble at the start of the century, has grown over the past few months. Investors are asking whether the billions of dollars spent by the likes of Google, OpenAI, and Meta to build even larger LLMs are justified and what the timelines of the returns on such investment are.

"You look at OpenAI, Anthropic, or Google and then you look at all the industry that feeds into it because these large-language models are ever larger," says Hodjat. "They need much more data and strong data centres to be able to train as well as inference."

That bubble, however, will not affect the adoption of agentification processes in enterprises because investment in those areas of AI enablement has been within reasonable limits, Hodjat added.

"Even if the bubbles were to burst to some extent next year or the year after, the enterprise adoption of AI wouldn't be tiny because of that."

Hodjat said the reason for that was multi-agentic systems would be required by enterprises in areas such as coding, human resources, marketing, and supply chain.

"The question is how I can best use AI to agentify my enterprise. I think we're making that transition and hope the year 2026 will be one in which we'll see more and more."

And yet enterprise adoption has been below expectations because organisations are figuring out their AI strategy, investment, and the proportionate returns within a reasonable span of time.

Despite widespread experimentation and investment, AI has struggled to translate early adoption into meaningful, scalable business value, exposing a gap between enthusiasm and impact.

The McKinsey Global Survey on the state of AI shows globally nearly 90 per cent of enterprises use it. Yet when asked how many fully scaled up those use cases, the number drops to 7 per cent.

Findings from the MIT Project NANDA report The GenAI Divide: State of AI in Business 2025, reinforce this disconnect. Despite an enterprise investment of $30 billion to $40 billion in generative AI, nearly 95 per cent of organisations are seeing no returns.

"People are recognising that AI enablement requires engineering. It requires proper design," said Hodjat.

"It requires tail layering and customising, and you can't just sprinkle your organisation with co-pilots and be done with it."