How Perplexity's founder thinks about AI, growth, and the future of work

Executive overview

Aravind Srinivas, CEO of Perplexity, built a company from a $150M seed valuation to $20B by obsessing over daily improvement and shipping fast. AI agents like Comet — Perplexity's new browser — are shifting power from advertising platforms to users for the first time. Entry-level jobs built on information retrieval are under pressure; the people who survive will push their value further up the stack.

The single most important trait is intellectual curiosity — loving to learn deeply about any topic.

Building Perplexity: the compounding growth mindset

• 1.01 to the power of 365 equals 37.78 — 1% daily improvement compounds to 3,700% annually.
• First action every morning: read all user feedback across social platforms and triage bugs immediately.
• Relentlessness is the company's defining culture — never giving up, constantly questioning how to improve.
• Brand partnerships (Formula One, Lewis Hamilton) are chosen to associate Perplexity with the greatest people of all time, not to drive trackable installs.
• Trust is earned through accuracy and speed of improvement; users who see the product meaningfully better week-over-week become advocates.

Comet: the agentic browser

• Comet can perform multi-step reasoning — find a video, pull its transcript, open the tab, start playback from a specific moment using natural language.
• It reads across tabs: ask it to summarise a one-hour interview and surface only non-trivial insights, then send that as an email directly.
• Recurring tasks are the next frontier: set a rule like "alert me and submit an application every time a house below $X appears in this neighbourhood."
• Businesses should focus on genuine product quality and organic reviews — Comet reads reviews across platforms, so cross-platform reputation matters more than any single channel.
• The model for ads: advertisers talk to agents, not users; agents can negotiate on the user's behalf and share revenue back with the user.

The future of search, shopping, and professional services

• Amazon and Google have always had AI — but it worked for advertisers, not users. Comet is the first AI that works exclusively for the user.
• In a world of AI agents, moats shift to physical infrastructure (warehouses, stores, supply chains), not the app or shopping experience.
• Financial advisors who only manage stock portfolios will be displaced; those who provide access to private equity, hedge funds, and off-market deals retain value.
• Real estate agents with off-market inventory remain useful; Redfin/Zillow-style listings become something an agent handles automatically.
• The ad margin compression is structural and permanent — user intelligence is now on the demand side for the first time.

On building in AI as a founder

• Building a foundation model is no longer viable for most startups — it requires a lab, a cluster, sustained talent, and a roadmap of iterative model releases.
• Product is more statistically probable than model-building, but every major lab and big-tech company is also building horizontal products.
• The only defensible bet is genuine obsession: build something you care about more than anyone else does, because well-resourced competitors will eventually enter every market that works.
• When an idea generates real revenue, expect the existing players to follow — design the company assuming that happens.
• The right strategy is not whiteboard market analysis. Build for yourself and bet that the problem resonates at scale.

On education, career, and the AI era

• A PhD's real value is learning how to learn — acquiring the ability to enter a new field, ask the right questions, and reach subject-matter depth quickly.
• At 18, committing deeply to one thing for one to two years matters more than knowing which field to choose.
• The confidence to take on hard problems comes from having done it at least once — without that track record, it is hard to bet on yourself again.
• For early-career workers, the question is not whether AI replaces tasks but whether you expand the value you deliver beyond those tasks.
• Wasting time on low-quality, distracted work does not count as working — uninterrupted focus on the product is where the real leverage is.

Aravind's personal working style and AI tool stack

• Works or thinks about work almost every waking hour; genuine deep-focus time during the day is when his best work happens.
• Primary tools: Perplexity, Google (for specific widgets and advanced search syntax), ChatGPT.
• Uses Midjourney occasionally for its aesthetic quality; tests all major AI apps as a competitive research requirement.
• Still uses Google for live scores, currency conversion, and finding very old links with advanced search operators.
• Comet changes this: it can orchestrate other tools (Midjourney, Photoshop) directly, removing the need to open separate apps.