This moment in AI is something that you just cannot miss: Perplexity’s Aravind Srinivas

Perplexity AI founder Aravind Srinivas.

Aravind Srinivas, founder of Perplexity, an AI-powered answer engine, spoke to News18 at last week’s Rising Bharat Summit for a conversation about the rapid advancements in artificial intelligence.

In an interview with Chandra R. Srikanth from Moneycontrol, Srinivas delved into the fast-evolving landscape of AI innovation, India’s potential to build a strong deep-tech ecosystem, and tweaking Perplexity’s pricing for the Indian market.

He also shared his perspective on the one piece of advice regarding artificial intelligence that he chooses to disregard. Perplexity, backed by prominent investors, has become a key player in the AI space.

Watch the entire conversation. 

EDITED EXCERPTS:

Give us a sense of the state of play in AI today. Every day, when you open your Twitter feed or any news website, it's just so overwhelming—the pace of creation and disruption. As someone who's sitting right there in the hot seat in the Valley, give us a sense of what's happening.

I think you're right. The last two to three months have been insane in terms of the speed of progress. I would say the industry has moved faster in the last three months for me than in my first two years of starting Perplexity. And that should say something, because I considered Perplexity to be fast-moving. I don’t think we've ever moved faster than we have in the last three months.

I believe this acceleration is partially driven by the entry of China, DeepSeek, and many other models from China.

Because of this, all these new capabilities are now possible, especially in reasoning, where the models can think through and answer more complex questions. These models can go through a long chain of thought before executing a complicated workflow, like doing the work of a research analyst or a professional researcher—performing background research, diligence research, or anything similar. All these tasks, which would have required sophisticated workflows before, can now be compressed into a long-running chain of thought with well-orchestrated LLMs, tools, data providers and APIs.

This is creating all kinds of new products, like deep research, which was never possible before. It’s also leading to the creation of things like agents. That’s why this year is so exciting.

Are you having sleepless nights? It’s taking just days for a rival to launch a competing product and out-execute an incumbent. So how do you keep up with that?

Well, there’s no substitute for hard work. So, we work hard. That’s why, after our conversation, I’m going back to work. This moment in AI is one you just can’t afford to miss. If you take a break and come back after three months, it’ll feel like the world has completely changed.

To make sense of what I mean—imagine it’s the end of December or early January. If I told you there would be DeepSeek, Deep Research, O3 Mini, all these different models, agents, image generation tools—everything that’s happened in just the first three months of the year—you’d probably say, “Come on, I know AI moves fast, but that sounds like too much.”

But that’s literally what happened. And if anything, we should expect things to move even faster for the rest of the year.

So, are you clocking 90 hours a week, Aravind?

I’m not sure. I try to work all my waking hours. There's this quote from Jensen Huang: "I wake up, I work. And even when I’m not working, I’m thinking about work." That pretty much sums it up. I don't think anyone in this space right now is doing anything different. It just seems like the bare minimum expectation right now.

Previously, every funding round for a startup coincided with a huge demand for engineers, full-stack developers, and mean-stack developers. But now we see a new trend where you can build 10x better AI products with very few engineers. What does this mean for the future of engineering jobs, especially in a country like India, which produces lakhs of engineers every year?

I would say that, at this point, if you're not using AI when you're writing code, you're essentially stagnating by sticking with the old workflow and tooling. People who use AI to write code are getting work done much faster. We see that internally at Perplexity, where some engineers are very good at using these AIs to write code, whether it's in their editor, like GitHub Copilot or Cursor, or in end-to-end deployment apps, like Replit or Bolt, or in separate standalone apps, like Perplexity or ChatGPT. They're able to fix bugs much faster.

So, I think the solution for engineers in India is to start using AIs as much as possible to write code and software. And then definitely, you'll need fewer people to do the same amount of work.

That doesn't mean that you don't need other people, but you're just gonna find a lot more projects. And so, we can get, like, a lot more software projects done.

And then it could also mean that, like, people with a lot of creative taste no longer need to feel like code monkeys if they want to, like, just get paid for something, right? They have an AI to do all that for them now, so they can truly be liberated, have all the freedom to go push new ideas into production themselves, own the users, build businesses on their own, potentially get to, like, million dollars in revenue, and recurring with just, like, a one-person company or two-person company. And that can lead to a completely different kind of economy, too.

I see endless possibilities. At the same time, I won’t say everything is looking good. There will be a transition period, and those who don't adapt will face difficulties.

Aravind, you spoke about what engineers should be doing, but what do you think all of us should be doing? Anyone who wants to experiment with AI, learn about AI, use it in their daily lives, how should they go about it? Where should they start?

I think we should all start learning from everything. This also applies to me. It doesn't mean that people running AI companies know how to use AI better. In fact, I still check emails in the same way I used to, and I'm still managing meetings and calendar appointments in the same way as before. And now I've been thinking, okay, well, you know, “shouldn't you ideally be asking an AI to do these things for you?”

For example, preparing ahead of a meeting, getting bios in advance—now you can enter anyone's name on Perplexity and get a solid bio. You can even enter five people’s names together, and it will still do that for you. AI can automatically annotate your day ahead of time, helping you be more prepared.

Similarly, when I'm trying to ask new questions about a topic I don't understand, whatever is going on in the global economy right now, I'm not that financially or economically literate. So, I can learn about a new topic as soon as something happens, without needing access to any expert. And I feel like that's a completely different future we're living in.

And then you can also customize it to your profession if you're a product manager or like a business development professional, coming up with strategies to get better at one particular task you're doing.

Commerce Minister Piyush Goyal recently compared India to China, highlighting that while India has a large startup ecosystem, it lacks innovation in areas like semiconductors and EVs. Given your background in IIT Madras, Berkeley, and building Perplexity in the US, what do you think India needs to do to create world-beating products in AI and technology?

Certainly, I think we can. Just to comment on the quote, I believe it's a very positively provocative statement. Generally, such comments are meant to provoke a reaction, like "Let me come and show that I can do it." I hope that's what it leads to.

To clarify, China went through a similar trajectory. Whatever India is doing right now—trying to build clones of apps that already exist but tailored for the Indian market—China did that for a long time. That's essentially what happened for a while, and they built the muscle to ship products quickly and iterate.

Only after that did they begin building new products that didn’t exist and were able to capture a global audience beyond the Chinese market. TikTok is a great example of that. So, India can certainly produce its own version of something like TikTok, which, while not deep tech, is still a product that could capture a global audience, not just the Indian one. It's definitely possible.

The venture capital scene is also crucial. I think VCs in India are still very risk-averse. They prefer investing in things they won’t lose money on, rather than betting on opportunities that could potentially yield huge returns, even if some of their investments fail.

So, the mindset should be like, okay, 95% of my investments would fail, but 5% of my investments would return to me like a hundred or a thousand X that the 95% that failed doesn't matter. I think that mindset doesn't quite exist in Indian venture capital right now.

You announced hiring plans to expand your operations in India a few months ago. How is that progressing? You're also focusing a lot on cricket content lately. Has that helped?

So, India is in the top 10 countries for us right now, probably even close to the top five in terms of revenue already. I'm not even talking about usage, just even revenue, which is pleasantly surprising because our product is priced pretty expensively right now.

A lot of people are using the deep research agent that we have. Our pro search is being used by a lot of people too. And it’s, in terms of daily query traffic too, the top five or top four for us. So, it's an important market.

In terms of enterprise, like knowledge worker adoption, India's quite far ahead of even countries like the United Kingdom, according to some research that's been done.

As for our hiring, I think it's honestly gone slower than I would have liked. We haven't hired that one person to really grow our business in India yet. We've got tons of applications. I went through some of them, but to be fair, we couldn't go through the entire list.

And I would have to spend a little more time on this because the expectation was only people with some amount of experience like general manager or like business development would apply. But basically, the interest was so big that a wide array of people with different sets of skills applied and so many referrals from people I know as well.

So, it got overwhelming to a point where I decided I would wait for a while and make sense of what we really need there before expanding.

And yes, focusing on cricket is very important. There are lots of searches on cricket that are already happening on Perplexity and we want to keep growing that. And we plan to also incorporate other content that's interesting to Indian audiences, like entertainment, TV shows, movies, all local results, better support for languages. All these things need to be there for a product to be vitally useful to the Indian audience.

India-specific pricing coming?

I don't know the number yet, but we're definitely thinking about it.

Many other consumer subscription apps have done this, like Netflix, Amazon. So that’s the way to go. Obviously, our inference is expensive. We don't want to put a lower quality product just to save costs there. So, we're trying to figure out a way to distill some of the open-source models, make it more efficient. And what should be in the lower tier plan and what should be in the higher tier plan? We're trying to think about it.

What's the one piece of advice in the AI industry that you deliberately ignore?

One is that people often want to plan, right? People want to have a strategy and a plan for one year or two years and stick to it no matter what happens in the world. I think that used to be a fine strategy when the world was moving a lot slower. It doesn't hold true anymore.

And so, I don't actually plan the company for more than three months right now. And this is not the best way to run any company, but in AI it happens to be like a perfect strategy for us.

The problem with AI though is you're building with a generic piece of technology. It's so hard to bake custom constraints into it. And the more custom you try to make, the more it loses its magical generality. And so without actually having an end user or like a demographic or vertical in mind, it's so hard to actually build anything as a product builder but you kind of have to figure out the sweet spot of still being useful for a particular set of users but making the product as general and horizontal as possible. And I think these are the two main reasons I would say a lot of people struggle to build AI native products in companies but it's essential.