The News
When CEO Aravind Srinivas launched Perplexity’s Comet web browser last month, he knew it wouldn’t live up to his vision of an operating system for the AI era, handling monotonous, time-sucking online tasks without any human help.
The AI models that power Comet weren’t good enough yet for that job and compute costs were astronomical. But that meant it was the perfect time to launch it, Srinivas said, reflecting go-to-market strategies in the latest tech boom.
“You’ve got to position your product and your technology with the assumption that the models are eventually going to be great and also going to be affordable,” says Srinivas, who founded his AI-powered online search startup in 2022.
Comet is different from other web browsers because of its right-hand column, where users can instruct an AI model to complete tasks, like going through LinkedIn profiles to find job candidates, filling out summer camp signup forms, or watching a YouTube video and summarizing it in bullet points.
But Comet, which comes with a $200 monthly Perplexity Max subscription, is still too limited to fulfill the goal of becoming like an operating system. It struggles with more complex, multi-step functions that require a lot of memory. And the vast majority of websites have not implemented AI functionality, making it difficult and resource-intensive for AI models to navigate.
On the tasks it can do, it delights its early adopters, who use it to send emails, add calendar entries, and place online orders, to cite a few uses. On X, Comet users post ideas for new ways to use the browser.
Srinivas says users are increasingly opting to make it their default browser and the waitlist has grown to more than one million. “The amount of people who are interested in trying and living, with all the imperfections, is pretty high,” he says.
Perplexity’s pace of innovation embodies the unique characteristics of product development in the era of generative AI, where instead of trying to perfect something, companies must anticipate what will be possible when yet-to-be-built AI models reach the market. That means walking a tightrope between launching a product that is too early and doesn’t work at all, or one that works perfectly but has already lost to fast-moving startups who got there first.
“It changes the way you think about building products,” said OpenAI Chief Product Officer Kevin Weil on a recent podcast with investors Bill Gurley and Brad Gerstner. “If you think about it right, it makes you much more open to building products that only kind of work. Because if the model can only kind of do it, then it’s going to be great in a few months.”
In this article:
Step Back
While earning his PhD in computer science at UC Berkeley, Srinivas wrote AI research papers that earned him accolades among his peers. But he chose to leave academia, and eventually landed an internship at OpenAI.
While there, OpenAI Co-founder Ilya Sutskever told Srinivas there were probably only two areas where you could work on AI and build a product at the same time: online search and self-driving.
The reason was that they were both part of an AI “flywheel,” where developing the product creates more data, which helps improve the underlying technology that powers it. And as the product gets better, it attracts more users, which in turn creates more data.
Both of those categories also get better as AI improves. So instead of getting run over by disruption, they could ride the AI wave.
The conversation helped push Srinivas to co-found Perplexity.
Being a startup CEO was a learned skill for Srinivas, who grew up in India and studied electrical engineering before coming to the US.
But the 31-year-old has learned quickly. “Business does not mean you wine and dine, do a bunch of charismatic statements, and do the Steve Jobs marketing,” he says. “You just have to be yourself. It’s still, at the end of the day, problem solving.”
Reed’s view
The “be yourself” mantra only partially applies to X, where Srinivas has honed an ability to post with a mix of wit, edginess and founder energy that he doesn’t display in everyday life. It’s a skill he says he learned in part from investor Marc Andreessen. “Marc told me you’re never supposed to tweet something that’s 100% correct because nobody cares,” he says. “The highest engagement is always tweets that are 50% correct and 50% wrong.”
Perplexity has been on a fundraising tear, raising $500 million in the spring at an $18 billion valuation.
The company has also rapidly iterated, adding features like deep research, image generation and shopping to its search engine.
The company’s success has drawn acquisition interest from companies like Apple and Meta. But Srinivas says he isn’t selling. “We want to ensure that little tech wins and we want to keep going. It can’t be more definitive than ‘we’re not interested,’” he said.
Perplexity’s move from search into a browser may seem like a departure from Sutskever’s advice, as it is neither search nor self-driving cars. But Srinivas doesn’t look at it that way. “It’s probably like a third category, which is a merge of search and self driving — agents on the browser. So, the browser feels like a car. Agents feel like autopilot,” he says. “I think after Comet, people clearly understand where we’re heading.”
Room for Disagreement
There’s a debate over whether Perplexity’s technology constitutes a novel use of AI agents or should be considered a “crawler” when it accesses outside websites. Earlier this week, Cloudflare published an investigation showing that when users asked Perplexity to access a specific website, it would do it, even if the website requested it not be crawled by the company.
Perplexity has said that if users type in a specific URL or website, it’s different from indiscriminate crawling.
But some websites say they want to be able to choose which companies are able to access their content. Microsoft-owned LinkedIn, for instance, could say it only wants OpenAI’s models to be able to access its pages, and not competitors like Perplexity.
Notable
- OpenAI is also planning to release a new web browser, according to Reuters, in a challenge to Google Chrome. The move shows how the agentic web browser could be the next battlefield in the AI wars.