Parag Agrawal says his AI agents search about 1,000 times more than he did
Agrawal bets his own agents do 100 to 1,000 times his old daily searches, with company use even higher.
Training Data · Parallel’s Parag Agrawal: Building a New Web for AI AgentsParag Agrawal, the former Twitter chief executive who now runs the search startup Parallel Web Systems, guesses he did 20 or 30 Google searches a day three years ago. He bets his own AI agents now do 100 to 1,000 times that, and counting company-wide activity might easily exceed a thousandfold. He adds that he is an outlier and agent adoption is still very early.
One prompt explains the jump. He says even a simple request to an AI app triggers five to ten web searches, because it can, while deeper research modes trigger hundreds to thousands. Background agents add more, since they watch the web and act when something changes.
To show the larger multipliers, he offers a hypothetical lender with 10,000 small businesses that moves a monthly human risk check to a weekly agent run, multiplying searches by hundreds of thousands to a million.
He recalled seeing Cloudflare monitoring showing AI page reads about equal to human reads, though he noted that includes crawlers and other automated traffic.
Questions people ask
Do hundreds of searches actually give a better answer?
It can, because an agent can check more sources, compare numbers, and retry when a page is missing. Agrawal says the limit is practical: more searching means more compute cost and delay, so developers deploy agents only where the incremental value justifies the spend.
Who pays for all that searching?
The app developer does, usually by paying for search, computing power, and data access. That is why many AI products have usage tiers, with deeper research modes costing more or being limited to paying customers.
Has AI web traffic already passed human traffic?
Not for searches, in Agrawal's view. He says agentic queries have not yet passed human queries and that adoption is very early. He recalled Cloudflare data showing AI page reads about equal to human reads, while noting that comparison includes crawlers and other automated traffic.