AI Agents Will Become Decision Makers In 8-12 Months: Ashish Khushu of L&T Technology Services

As the buzz around generative AI grows louder, enterprises are grappling with both the opportunities and risks of this transformative technology. At the Nasscom Generative AI Foundry Bootcamp, Outlook Business caught up with Ashish Khushu, CTO of L&T, for an in-depth conversation. In this candid Q&A, he explains why AI is leveling the playing field for companies big and small, how Agentic AI will redefine automation in the near future, and why India must balance building sovereign language models with practical applications.

How do you see AI changing the competitive landscape for enterprises today?

In AI, everyone begins on an equal footing, that’s the beauty as also the risk. AI provided an equal opportunity for startups and enterprises across the globe to get onto the bandwagon of experimentation, learning and contributing to its evolution and maturity. While it presents an opportunity to many to catapult themselves into using AI for competitive advantage, it could also have severe consequences if we don’t factor it into our investment plans or technology roadmaps over the next few years.

The adoption of digital for many companies was a default answer “Moving to the Cloud.” That obsession with infrastructure centric strategy lasted for a long time. With the advent of AI, I think people have quickly learnt that it’s going to be about use cases. A focus on business problems will lead to the broader adoption of AI and that’s good. Technology infrastructure is important and critical but follows the requirements of what we are trying to solve.

Generative AI and agentic AI are the buzzwords today. Where do you see them heading in the next couple of years?

Agentic AI is especially exciting. It allows you to build “agent farms” that can automate discrete tasks at scale. Today, creating agents is relatively simple, but the real leap will come when reasoning capabilities mature. In the next 8-12 months, I expect reasoning to become much stronger, enabling systems to take autonomous decisions. That’s when AI will move from being just a tool to being a decision-maker, unlocking massive automation opportunities.

Hallucinations remain a sticking point with AI models. Has the problem improved?

It has improved, but it hasn’t disappeared. Hallucinations happen when models generate inaccurate or misleading results, often due to the quality of training data or the way the model is built. Today, we know more about how to mitigate it, but as AI systems become more complex, hallucinations take different forms. The long-term fix lies in better-quality data and stronger algorithms and stronger audits of results. So yes, it’s better than before, but the problem isn’t solved yet.

India is multilingual, yet we often hear about the lack of data for Indian languages like Hindi. How do you approach that challenge?

That’s a big issue. Generally speaking, the availability of high-quality local language databases for different applications is very limited compared to English. This affects not only start-ups but also large enterprises. The only real solution is collaboration with customers or agencies that already hold relevant data.  By all accounts, the realm of AI is expanding on all fronts, so a lot of work is happening in all of these areas and communities contributing to these advancements are increasing by the day and thereby enabling quicker maturity. It’s a matter of time when all of these things will be behind us.

There’s debate about whether India should build its own sovereign large language models or focus on applications. What’s your view?

I don’t see it as an either-or. Both approaches can and should coexist. Building sovereign language models requires deep pockets, patience and long-term commitment. Application development, on the other hand, delivers faster business value. Whether India invests in sovereign models depends on strategic choices by the government or academia, but eventually, they will emerge as part of organic growth. I see this as inevitable rather than optional.

Finally, should enterprises view AI as a disruption to be wary of, or as a tool to embrace?

AI should be seen as a very effective tool. Nothing more, nothing less. It’s powerful, yes, and it will transform the way we work, but it’s not something to be scared of. The power and impact of AI will be determined by the technology strategy enterprises have. We surely shouldn’t treat AI with awe or fear, it’s simply a new technology in our toolbox, one with immense potential but here to stay and is already becoming an integral part of our life.