In 2025, developing AI applications becomes easier than ever, but survival for startups becomes increasingly difficult.
Not because the technical threshold is too high, but because giants like OpenAI and Google are crushing every emerging innovator with capital and traffic. When ChatGPT can generate code at the click of a button, and Gemini can automatically complete business plans, what can startups rely on to establish themselves?
Recently, at a Y Combinator startup sharing session, Perplexity CEO Aravind Srinivas rarely opened up a wound, showing how an AI startup can grow wildly in the cracks between Google and OpenAI.
This 40-minute dense conversation is not the usual success story but a survival guide for AI entrepreneurship:
- Speed is life: when giants start to copy your ideas, your product iterations must be faster than their decision-making processes
- Precise positioning: establish a "cognitive moat" in niche areas where giants cannot fully commit
- Embrace defects: turn every bug into an opportunity for co-creation with users
This article will analyze how Perplexity broke through the giants' blockade in three years, distilling the three major iron laws of entrepreneurship in the AI era: pinpoint breakthroughs × extreme speed × user co-creation.
01 Find the battlefield where giants "cannot fully commit"
"Google has the best engineers in the world, but they just cannot do AI search well," Aravind bluntly stated on stage.
When the entire audience gasped at this "provocation," he followed up with a sharp observation: the giants' greatest advantage is precisely their biggest weakness.
The 2023 Google Bard demonstration failure led to a 6% drop in stock prices, becoming a classic case in the AI era.
Aravind's interpretation of this is refreshing: "It's like asking an Olympic gymnastics champion to also compete in weightlifting. Google's advertising machine generates $200 billion in revenue each year, but this also means:
1. They do not dare to directly integrate AI answers into core search (it would disrupt the advertising ecosystem)
2. They cannot afford any public mistakes (stock price sensitive)
3. They will not rebuild business models for AI (the opportunity cost is too high)
This dilemma of "dancing with shackles" gave Perplexity a precise breakthrough point:
"We specialize in what Google cannot do—providing precise answers without ads and indicating each information source."
A detail from the live demonstration is highly persuasive: when a user searched for "What scenic hotels are near the Golden Gate Bridge in San Francisco?" Google prioritized showing ads from bidding platforms, while Perplexity directly listed hotels that met the criteria, along with real ratings from Tripadvisor and booking links.
"This is what search should look like," a serial entrepreneur commented from the audience, "Google knows the right answers, but their business model does not allow them to show it."
02 When AI can write code, what is a true barrier?
"Now fixing a bug is faster than writing a new feature used to be."
Aravind's demonstration of the R&D process left traditional developers speechless:
▪ Product managers take photos of UI issues with their phones and directly import them into Cursor AI
▪ AI automatically generates SwiftUI code modification suggestions
▪ Engineers review and immediately push hot updates
For example: In March of this year, users reported "unable to save long conversation records." Traditional companies might need:
1 day to reproduce the problem
3 days to discuss the solution
1 week development testing
And the Perplexity team:
10:00 AM: User email arrives at CEO's inbox
10:15 AM: AI automatically generates 3 solutions
11:30 AM: Select solution and complete coding
2:00 PM: Update pushed to 20% of users
Before 5 PM: Full release based on feedback
"This is not a technological miracle, but a cognitive reconstruction," Aravind explains, "We no longer pursue 'getting it right the first time,' but rather 'fast trial and error.' Fixing 20 bugs a day is more important than releasing one perfect version a week."
The compounding effect brought by this speed is astonishing:
- User retention rate increases threefold (because demand response is measured in hours)
- Employee productivity increases fivefold (AI handles 70% of repetitive coding)
- Product iteration speed surpasses Google’s similar feature update cycle
"Google IO releases 'AI search new model' once a year, we have a major update every two weeks," Aravind's timeline comparison elicited knowing smiles from the audience.
03 Browser: The forced "Normandy Landing"
"If we only do search, we will eventually be swallowed by ChatGPT."
When Aravind announced All in Browser, even YC partners showed surprise.
This seemingly risky decision is backed by a cruel survival arithmetic:
Countdown to the death of traditional search
"The browser is our D-Day," Aravind uses a military metaphor to explain this all-or-nothing decision: "When all shores are blocked by the enemy, you must create your own landing point."
Cognitive operating system vs chat box
The killer scenario of the Perplexity browser is eye-opening:
🔺Scenario 1: Automatic Price Comparison
"Find the cheapest San Francisco-London flight in the past six months, excluding red-eye flights" → Automatically scan 10+ websites to generate a price comparison report
🔺Scenario 2: Research Assistant
"Organize the funding cases in the AI pharmaceutical field over the past three years" → Parallel scan Crunchbase/academic papers/financial call records
🔺Scenario 3: Personal Butler
"Using my calendar and email data, find the 3 best time slots for working out next week" → Automatically book the gym
"This is no longer a tool, but an extension of the cerebral cortex," Aravind demonstrated a browser running 12 asynchronous tasks simultaneously—including monitoring competitor product updates, automatically renewing cloud services, and tracking logistics—just like background processes on a computer, but all interacted with in natural language.
04 The new role of the CEO: Chief Bug Officer
"This morning I personally fixed 3 bugs, which may be the worst case of CEO time management ever."
Aravind's self-deprecation resonated with entrepreneurs.
In this era where AI reconstructs everything, the role of leaders is subtly changing:
Traditional CEO
- Strategic planning
- Fundraising roadshow
- Team building
CEO in the AI era
- Product sensitivity radar: can detect bad smells in the code
- User feedback decoder: extract demand signals from complaints
- Defect excavator: turn every bug into an improvement opportunity
"Kitchen Theory" management method
Aravind's shared team culture is impressive:
"I treat the company as a restaurant kitchen, and all engineers must rotate through customer service. When you see users miss flights due to search errors, that shock is more effective than any KPI."
This kind of "immersive management" yields astonishing results:
- Customer service response time reduced from 6 hours to 23 minutes
- Engineers voluntarily organized "Bug Fix Marathon" fixing over 100 edge cases weekly
- User emails directly drive 60% of product improvements
"Google might never be able to do this," a former Google engineer whispered from the audience, "Their CEO cannot see the real pain of users."
05 The overtaking curve of small companies
"AI hasn’t changed the essence of business; it has just reduced the cost of innovation from millions of dollars to the price of a lunch."
——Aravind Srinivas closing remarks
The most shocking moment of the speech was a comparison image shown by Aravind:
Asymmetrical resource wars
"This is the true dividend of startups," Aravind's summary resonates strongly: "When giants must fight with aircraft carriers, you can land in their blind spots with a speedboat."
As they leave, each audience member receives a special "gift"—a card printed with a list of all publicly known bugs of Perplexity. "Welcome to challenge our weaknesses," this Zen-like gesture may be the most seductive entrepreneurial declaration of the AI era.