Mixed Signals · Nov 6, 2026 · 1h 1m · 52m saved
With Parag Agrawal, Founder of Parallel and former CEO of Twitter
Photo: icpcnews icpcnews, CC BY 2.0 via Wikimedia Commons
Parag Agrawal founded Parallel three years ago on a bet that AI agents will eventually query the web roughly 1,000 times more than humans do. Most AI products today disguise an agent as a human browser clicking through a website; Parag says that disguise is exactly what makes agent search slow and expensive.
"If this doesn't happen, then I don't get to transact. If I don't get to transact, I can build all the info I want, but without the information and the data, my business is done." — Parag Agrawal
Parag ran Twitter as chief executive until Elon Musk's 2022 takeover, and he told Mixed Signals the business problem he faced there was never solved. He now spends his time arguing that the web's ad and subscription models cannot survive agent traffic unless publishers get paid directly.
The full interview is covered here so you can skip it. 61 minutes of audio, 9 minutes of reading.
Here are the 7 calls that matter.
Key Takeaways
Parallel was built on a bet that agents will query the web 1,000 times more than humans
The limit on human use is time and attention; the limit on agent use is only whether the resulting work pays for itself
Agent search needs to be reinvented, not adapted from human search
A real-time voice agent needs an answer in about a second; an overnight agent can take far longer and go deeper
Planning one trip to Slovenia used 15 to 50 times more searches than Parag would have run himself
His agents also got blocked on many sites along the way
Ads and subscriptions both break once there is no human eyeball or wallet on the other end
Publishers' rational response is to block agents outright, which Parag calls "value-destructive"
Parag wants content priced by which AI is reading it, not by a flat subscription
A frontier model costs 50 to 100 times more to run than a smaller one for the same task, and he wants content priced the same way
Parallel already pays publishers including Fortune and The Atlantic, in a coalition Parag calls a "Rebel Alliance"
He says direct deals between publishers and the big labs get worse even as the labs' usage grows 5 to 10 times a year
Twitter the product outlasted Twitter the business, Parag says
He still hasn't seen anyone solve the business problem he left behind
1. A Bet On 1,000x More Web Use
Three years ago, before agents existed as a product category, Parag and his co-founders built Parallel on a single hypothesis: AI agents will eventually use the web roughly a thousand times more than people do. Human web use is capped by how much time and attention a person has; agent use is capped only by whether the resulting work is worth paying for.
Parag says agent demand has no natural ceiling the way human demand does
"It was with this hypothesis that agents will use the web a thousand-x more than humans... three years ago we were walking around, there were no agents, telling people that agents will use the web a thousand-x more than humans ever have." — Parag Agrawal
That scale, he argues, is not a bigger version of the same problem. It requires search infrastructure built from scratch.
"If you sit down and think about it for a minute, it is somewhat ridiculous... you need to invent multiple generations of technology in order to serve a need at that massive scale." — Parag Agrawal
