By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.
First Round CapitalTwitter's former CEO on rebuilding the web for AI | Parag Agrawal (Co-founder and CEO of Parallel)
Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.
Great conversations run long. Their best ideas don't have to.
uListen is free on iPhone, iPad, Android, Chrome, and Claude.
See howAt a glance
WHAT IT’S REALLY ABOUT
Parag Agrawal explains rebuilding the open web for AI agents
- Parag Agrawal says Parallel builds APIs that let AIs access the web, based on the belief that AIs will use the web a thousand to a million times more than humans, and that its products hit human-level or better performance on deep research benchmarks.
- Parallel designs for an AI customer rather than a human one, which let it drop latency constraints and focus on accuracy, quality and slow end-to-end automation of repetitive web research work.
- Agrawal describes moving from Twitter CTO/CEO to founder as an exercise in unlearning, covering in-person work, stochastic AI products, shaping a company around yourself, and harnessing public attention.
- On company-building, he describes a hiring approach that bets on high-potential people and is willing to fire, a flat 22-person team with no managers, a fundraising rule tied to binary odds of success, and uneven results from AI coding tools.
- He sees the main open challenges as measuring AI quality through evals and keeping web data open, and proposes attribution, differential pricing and open markets as the basis for incentives.
IDEAS WORTH REMEMBERING
5 ideasDesigning for an AI customer changes the product assumptions
Agrawal says humans have a second or two of patience and underspecify queries, while AIs can specify goals and tolerate slow responses, so Parallel chose to optimize for quality over latency.
Early use cases came from repetitive human outsourced work
Parallel went after desk research like insurance claims, underwriting and financial data work done by BPOs and KPOs, because AI performance improves on well-specified repetitive tasks while human performance declines at scale.
Ground truth is never ground truth
After scoring low against a customer's human-generated dataset, Parallel learned to compare two approaches, look only at disagreements, and grade a small sample to judge which is better.
Keeping the early customer set small preserves flexibility
Following Patrick Collison's advice that the first API design will be wrong, Parallel limited API access so it could change versions even in backward-incompatible ways.
Raise only what you can argue improves binary odds
Agrawal treats building a company as success or failure, so he raised up to the point where he could still articulate how more money raised his odds, and no further.
WORDS WORTH SAVING
5 quotesWe believe that AIs will use the web a thousand X, a million X more than humans ever have.
— Parag Agrawal
So in some sense, this startup is like a journey in unlearning everything I might have learned before.
— Parag Agrawal
Turns out ground truth is never ground truth, right?
— Parag Agrawal
To me, building the company is binary. Either we are successful or we're not. There's just, like, nothing in the middle, right?
— Parag Agrawal
You're constantly under-qualified to be what your aspiration, ambition, and company needs from you, right?
— Parag Agrawal
Ask ChatGPT about this episode
Get its key ideas, plus a summary of each chapter linked to that point in the video, right in your chat.
Add uListen AI to ChatGPT.
Install uListen AI in ChatGPT (opens in a new tab)Needs a ChatGPT account.Paste this prompt into a new chat.
Use uListen AI to give me this episode's key ideas and a chapter-by-chapter summary, with timestamps: https://www.youtube.com/watch?v=fORbmCMX5GA
SPEAKERS
Parag Agrawal
guestCo-founder and CEO of Parallel and former CEO of Twitter.
Todd Jackson
hostPartner at First Round Capital and podcast host/interviewer.
High quality AI-generated summary created from speaker-labeled transcript.
