By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.

First Round Capital

Twitter's former CEO on rebuilding the web for AI | Parag Agrawal (Co-founder and CEO of Parallel)

Parag Agrawal is the co-founder and CEO of Parallel, a startup building search infrastructure for the web’s second user: AIs.

Parag AgrawalguestTodd Jacksonhost

Aug 14, 20251h 5mWatch on YouTube

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.

CHAPTERS

Jump to publisher timestamps

Chapter summaries

What Parallel Does and Why It's Been in Stealth
Parag Agrawal explains that Parallel builds tools AIs can use to access the open web, on the belief that AIs will use the web far more than humans. He says the company has been quietly building with forward-looking customers and now feels its product is ready.
- •AIs will use the web a thousand X to a million X more than humans
- •Parallel's recent APIs hit human-level and higher performance on deep research benchmarks
- •Seed round in 2024; company is under 18 months old
- •Early customers shaped the product, which is why the company stayed quiet
Unlearning Twitter: Building Product for Stochastic Systems
Todd Jackson asks about moving from a decade-plus at scale at Twitter to a clean sheet in a post-ChatGPT world. Parag describes how team structure and product thinking differ, and explains Parallel's bet on an AI customer and a deliberately slow, quality-first API.
- •Startup life is a journey in unlearning, from team size to practices
- •Twitter was remote-friendly; Parallel is in person five days a week
- •Product thinking now means building stochastic rather than deterministic systems
- •Designing for an AI customer means latency can be relaxed in favor of deeper work
- •Parallel bet on models getting better to support end-to-end automation
From CTO to CEO: Shaping the Company Around Yourself
Parag reflects on the shift from CTO to CEO at Twitter, where he decided to shape the company around his beliefs rather than fit himself to it. He describes the changes he started, including leadership and structure changes.
- •As CTO he shaped himself to the company's needs; as CEO he chose to shape the company to him
- •Changed the leadership team substantially within a month
- •Planned changes to headcount and roadmap that he did not get to see through
Public Attention and Twitter's Unfinished Transformation
Parag discusses how he views public attention as valuable for a founder, though he couldn't speak during the Twitter era. He shares an anecdote about a planned trim of 20-25% before the earnings call and comments on Community Notes.
- •Public attention helps find mission-aligned people, customers and partners
- •During the Twitter era it felt zero-sum, now it feels positive sum
- •A plan to trim the company by 20-25% was set for the earnings call days after the deal was signed
- •He is glad Bird Watch lives on as Community Notes
Twitter's Ethos Carried into Parallel and the Decision to Found
Parag describes the permissionless, open-access ethos he carried from Twitter into Parallel. He explains how, after leaving, he realized he couldn't simply join someone else's mission and decided to start a company, with advice from Josh not to rush.
- •Twitter's core ethos was permissionless access to everyone's ideas
- •Parallel wants an open, permissionless web that AIs can fully access
- •He knew within a couple of months he would start something
- •Josh advised him not to rush into the first thing
How the Idea Emerged: The Web for AI Agents
Parag recounts starting in healthcare agents but becoming obsessed with the future of the web. He contrasts human browsing, with its short patience and underspecified queries, against AIs that can specify goals, and explains the narrow slice Parallel chose first.
- •Idea began as science fiction: the primary consumer of the web becomes an agent
- •Humans have a second or two of patience and underspecify queries
- •AIs can specify end goals, enabling varied latency, volume and format needs
- •Parallel chose slow, measurable, repetitive, eval-friendly work first
Early Use Cases, Customers and Ground Truth
Parag describes the first use cases of BPO/KPO-style repetitive web research and the breadth of today's 100+ customers. He tells how an insurance design partner taught him that ground truth is never ground truth and led to a comparative evaluation approach.
- •Early use cases replaced repetitive outsourced desk research
- •Requests range from under half a cent to $10 and run to millions per day
- •Use cases span long deep research to three-second search for coding agents
- •Lesson: ground truth is never ground truth, so compare two alternatives and grade disagreements
Clay as a Design Partner and the Ideal Customer
Parag explains how Clay co-created benchmarks with Parallel and uses its Task API behind an AI orchestration layer, so an AI is querying Parallel. He then describes a broad customer base, alternatives like BPOs and plain OpenAI search, and Parallel's quality-first value proposition.
- •Clay built benchmarks with Parallel and pushed its accuracy
- •Clay's system turns underspecified user prompts into precise queries to Parallel's API
- •Customers include coding agents, insurance claims workflows and some BPOs
- •Alternatives are BPOs and build-your-own or OpenAI search
- •Value prop is quality first, then price performance
Competition, Launch Timing and Fundraising Framework
Parag says he doesn't think much about competitor dynamics because he sees a positive-sum future and fast-moving opportunity. He shares Patrick Collison's advice that a first API design will be wrong, and his rubric for choosing investors and sizing a raise.
- •Hyperscalers and labs can't move as fast as a startup; differentiation is a product decision
- •Patrick Collison's advice: your first API design will be wrong, so keep the customer set small
- •Chose investors by whether disagreement would keep him up at night
- •Raise only the amount where he can argue it increases binary odds of success
Hiring, Team Structure and Firing
Parag outlines a hiring philosophy of betting on potential while pairing high-potential people with those who can channel them, plus being willing to fire. He describes a flat 22-person team without managers or formal sub-teams.
- •Hire to maximize upside, betting on potential over experience
- •Include people skilled at channeling high-potential colleagues
- •Betting on upside requires being decent at firing
- •No formal teams or managers, with rotating project groups
- •Expect more to be done by fewer people, though opportunities pull toward hiring
Working with AI Coding Tools and the Engineer's Role
Parag admits vibe coding without understanding created cleanup work for his team, and says productive AI use is a learned skill. He describes giving AI access to CI and encouraging experimentation, and says what to build and taste will remain core, with declarative direction ahead.
- •Vibe coding without understanding wastes time and burdens reviewers
- •Giving AI access to CI makes the large crawler and index codebase friendlier to agents
- •Engineers get license to experiment rather than standardize
- •What to build and taste remain the engineer's role
- •Code is today's 'how'; the 'what' isn't yet captured in repos
Agents in Production: What Works, Evals and Defining Agents
Parag says elite customers excel at setting expectations with end users and operate in founder mode. He admits Parallel hasn't solved helping customers measure quality, discusses automating eval estimation, and gives his definition of agents.
- •Top customers communicate how humans should collaborate with AIs
- •Common thread among customers is founder-mode, forward-looking risk taking
- •Evals are hard because each scenario differs and keeps changing
- •Goal is a customer contract around quality
- •Agents are AIs with agency over how to solve a problem and access to tools
Parallel's Strengths and the Agentic Stack
Parag says Parallel is strongest in high-quality web research with fine-grained attribution and calibrated confidence scores. He lays out the top three tools every enterprise agent needs: web, internal data and a code sandbox.
- •Fine-grained source attribution enables confidence scoring on outputs
- •Knowing when answers may be wrong guides compute allocation and builds trust
- •Software will be expected to have LLM and web access
- •Top three agent tools: web, internal data, code sandbox
Keeping the Web Open: Incentives and Differential Pricing
Parag warns of an existential risk that the web closes up, and says incentives must make staying open better than paywalling. He proposes open market mechanics with differential pricing, and says Parallel wants to build an open marketplace with partners.
- •Risk that the web closes up unless incentives favor openness
- •A growing pie of AI value makes collaboration possible
- •Ads worked via differential pricing; content access needs similar mechanics
- •Attribution to sources helps with the math of feasibility
- •Plan to find partners and pay for or incentivize data sharing
Personal Obsessions and Growth as a Founder
Parag shares that he is geeking out on how the future web index will be built and sees recommendation, search and database systems converging. He says he does best when uncomfortable and names simplifying, communicating and expanding the coalition as areas to level up.
- •Reads research papers on the future web index
- •Believes recommendation, search and database systems will converge
- •Prefers jobs he wouldn't hire himself for
- •Needs to level up on simplifying and communicating, and expanding his coalition

Publisher timestamps

Get more out of YouTube videos.

Read along on your phone, search it beside the video, or ask your favorite AI app what was said.