Srinivas's YC talk is a wager statement: the browser — not the chatbot, not the protocol stack — is where agents become real, because it is the one surface that needs nobody else's cooperation. Around it, an operating doctrine stated without varnish: speed is the only moat, incumbents are caged by their own margins, and Perplexity's economics never need to be Google's to work.

The pivot thesis: search was the wedge, the browser is the bet — the only surface where an agent can act across the whole web without permission from anyone. And he scopes it deliberately down from the hype.

“We think about it as an assistant rather than an a complete autonomous agent.”

watch at 1:50

Why the incumbent can't follow: direct answers with booking links torch the ad machine, and the org that would have to accept that risk is a much bigger business, no longer founder-led.

“how are you gonna mint money from booking in Expedia”

watch at 12:40

“It's hard to take risks.”

watch at 15:02

His structural argument against protocol stacks: committing to MCP means betting on every third party's engineering, while a browser agent that navigates like a human needs nobody's cooperation — a bet this page's publisher takes the other side of.

“you require these third party MCP servers to work reliably”

watch at 27:13

“design it as the way a human would use that website”

watch at 27:29

The honest economics: nothing in history — Google's own second acts included — has matched Google's margins, and he does not claim Perplexity will. Subscriptions plus usage pricing at billions is the target, margin humility built in.

“even Google themselves never has had another business that had the margins that Google has”

watch at 28:05

Speed, demonstrated rather than preached: the answer engine shipped seven days after ChatGPT. The launch itself is his argument that velocity is the only defensible mode.

“We launched it seven days after the ChatGPT launch”

watch at 10:22