Why are enterprises still hesitant to adopt AI? Tech Mahindra's Kunal Purohit answers

AI has freed engineers in revolutionary ways, but there are speed breakers that stand in the way of the adoption of this new technology by enterprises.

This was the essence of the Visionary Address on 'The Vibe Coding Era: How India's Software Services Ecosystem is Reinventing Itself' that Kunal Purohit, the President – Next Gen Verticals Tech Mahindra, delivered at the 6th edition of Techspectations in Kochi on Friday.

Purohit said that the coder of the past had now, with the arrival of AI, become the "orchestrator of the present". The traditional coder focuses on syntax, manual line-by-line typing and how to write the code. Now, with AI in its adolescent stage, the coder is the co-creator. "In this stage, there is human-AI collaboration using co-pilots that accelerates the boilerplate delivery," Purohit said.

"The co-pilots or AI agents help you to write code and deliver outcomes you want," said Purohit, who drives AI adoption at Tech Mahindra. He said that Tech Mahindra at present has 30,000 employees working with close to 1000 AI agents.

From the co-creator springs the orchestrator, the third stage. Here, the coder of the past focuses on systems thinking, prompt engineering and rigorous verification of AI output.

Thus, a junior software engineer, who writes boilerplate code, becomes an 'AI-assisted product engineer' who designs features and validates AI-generated output. A 'senior developer' who focuses on module implementation becomes a 'system and agent architect' who designs AI-enabled system architecture. And a 'tech lead' who does sprint coordination becomes an 'AI delivery orchestrator' who does AI-human workflow orchestration.

The burgeoning increase in AI agents has created what Purohit calls the FAAFO advantage. FAAFO has five benefits. One, the development is 'fast'. "A lot of coding is quickened. The AI agents can be your assistant while you can focus on important aspects like what the intent should be," Purohit said.

Two, it can make enterprises 'ambitious'. "Because of the accumulated technical debt over the years, legacy software enterprises are not able to modernise. Now enterprises will have the time to indulge in ambition," he said.

Three, the work becomes 'autonomous'. "With codes being written autonomously, the workflow will be expedited," he said. Four, work will become 'fun'. "Since AI does the coding, one can use natural language to get things done. You can get a website ready in minutes with just the right prompts," Purohit said. In fact, vibe coding is about software development using natural language prompts.

Yet another advantage is 'optionality'. 'You have so many choices to pick from, various language models to choose from," Purohit said.

The FAAFO (Fast, Ambitious, Autonomous, Fun, and Optionality) model has generated a "tremendous amount of excitement". There is a general perception that production and turnaround times have gone up by 60-70 per cent.

Though potentially invigorating, the adoption has not taken place at a proportionate pace. Purohit identified five implications while adapting to an AI-based enterprise. One, AI needs to be grounded to the real context. "A lot of hard work is required to create knowledge," he said.

Two, the AI-based systems should be made more deterministic. "You cannot have a probabilistic scenario where the AI throws up outcomes that are random. Only with near deterministic outcomes can an enterprise inspire confidence," he said.

Three, the security aspect has to be worked out. "Many of us, on the consumer side, use open source software to create their own automation and workflows. But with this comes the risk of breach. We need to ensure that every single bit of work being done by AI, its role and authentication are clearly understood," Purohit said.

For instance, if an AI agent, with access to data, cannot be expected to function with the same moral discipline as a human support engineer. It could sell the data on the dark web, play mischief. Reimagining role and authentication, therefore, becomes critical.

Four, the enterprise has to continuously look at the quality of data. If the quality of data is poor, the outcomes would be bad. "Service providers should ensure that the model that runs in the data set is able to generate the right outcomes," Purohit said. And this has to be continually updated.

Five, change also has to be managed within the enterprise in a sustained manner. "This is the first time human beings are working with autonomous agents with a mind of their own," Purohit said. Perhaps because of all these factors, Purohit said that organisations were inherently struggling and not necessarily ready for such a shift.

Nonetheless, the transition to an AI-assisted world can be beneficial. "AI will compress repetitive execution and elevate engineering judgement," Purohit said. He said that 2026 should be seen as the "year intent becomes the primary interface for engineering". "The ideal should be to focus on intent. Other underlying things can be done by AI," he said.