Aug 14, 2025 · 1h 5m · in-depth
Twitter's former CEO on rebuilding the web for AI | Parag Agrawal (Co-founder and CEO of Parallel)
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Former Twitter CEO and Parallel co-founder Parag Agrawal joins First Round's Todd Jackson to discuss rebuilding web infrastructure specifically for autonomous AI agents. He shares key insights on unlearning big tech management practices, engineering for non-deterministic models, and establishing economic mechanisms to keep the web open.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
Brett as informed peer 4.7 Guest teaching 5.0 Guest disagreement 1.3 Brett pushing back 1.1
speaking balance: gold is Brett, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 9:13 CTO vs CEO Scope Pushback
Parag firmly rejects Todd's assumption that acting as CTO had the same transformative scope as taking over the CEO seat, explaining how structural change required full executive ownership.
Hardest push from Brett ▶ 34:35 Pressing on Hyperscaler MoatsTodd directly challenges Parag on how a startup can defend territory against well-capitalized hyperscalers and frontier AI labs.
Biggest teaching moment ▶ 1:00:15 Differential Pricing for Open Web SurvivalParag articulates an economic solution for web publishers using differential pricing mechanics to prevent the web from becoming closed and paywalled.
Brett holds their own ▶ 17:17 VC Picking Muscle AnalysisTodd brings his venture perspective to analyze how founders must develop the rare picking muscle to choose a decade-defining technical problem.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Brett as informed peer | Guest teaching | Guest disagreement | Brett pushing back | Why |
|---|---|---|---|---|---|---|
| Unlearning Big Tech Habits to Build an AI Startup | 5 | 4 | 1 | 1 | Todd leverages their shared history building the Twitter home timeline ranking to frame the shift from legacy scale to a fresh zero-to-one startup. Parag outlines how building with AI requires unlearning deterministic product frameworks in favor of stochastic systems. | |
| Designing Slow APIs: Prioritizing AI Accuracy Over Human Latency | 5 | 5 | 2 | 2 | Todd points out that performance latency matters when humans wait, prompting Parag to explain why Parallel intentionally designed an excruciatingly slow API optimized for AI background agents. Parag also politely corrects Todd's assumption that CTO leadership was identical to taking the CEO founder role. | |
| The Unfinished Twitter Transformation and Community Notes | 4 | 5 | 1 | 1 | Todd invites Parag to reflect on the post-Twitter acquisition drama and community notes. Parag shares the untold anecdote about his planned 25 percent company trimming and internal Project Saturn origin. | |
| Twitter Ethos and the Decision to Found Parallel | 5 | 4 | 1 | 1 | Todd shares VC insights on exercising the picking muscle when founding a company. Parag explains how Twitter's open permissionless ethos directly inspired Parallel's mission to keep the web open for AI agents. | |
| The Vision of an Agent-Centric Web Infrastructure | 5 | 6 | 1 | 1 | Todd prompts for the technical divergence between human and agent web interaction. Parag educates on how AI agents invert web browsing norms by eliminating narrow latency constraints and specifying precise end goals. | |
| Target Use Cases: Automating Complex Repetitive Workflows | 4 | 5 | 1 | 1 | Todd inquires about practical use cases and pricing metrics. Parag explains how automating repetitive human BPO workflows exposes the divergence between declining human consistency and improving AI performance. | |
| Early Design Partnerships and Quality Benchmarking with Clay | 6 | 5 | 2 | 1 | Todd references portfolio synergies with Clay, jokingly nudged by Parag who points out Todd made the intro. Parag details their technical partnership and explains why ground truth in human datasets is fundamentally flawed. | |
| Horizontal Product Breadth, Customer ICPs, and Competitive Dynamics | 5 | 5 | 2 | 1 | Todd presses on competitive moats against hyperscalers and frontier labs. Parag dismisses traditional competitive anxiety, arguing labs and hyperscalers remain too slow to capture every emerging niche in a positive-sum market. | |
| API Iteration Strategy and Capital Allocation Framework | 5 | 5 | 1 | 1 | Todd explores founder fundraising and API release timing. Parag lays out his binary framework for capital allocation and Patrick Collison's advice on planning for initial API design obsolescence. | |
| High-Alpha Hiring, Decisive Firing, and Flat Team Structure | 5 | 5 | 2 | 1 | Todd summarizes Parag's hiring strategy around high-potential talent vs experienced hands, and Parag refines the definition around managers who unlock high-alpha people while emphasizing decisive firing. | |
| AI-Assisted Engineering: Pitfalls of Vibe Coding and Declarative Systems | 4 | 5 | 1 | 1 | Todd asks about the realities of code generation. Parag candidly admits the perils of vibe coding in complex production codebases and argues APIs should be strictly declarative. | |
| Productionizing AI: Managing Evals and Defining Autonomous Agents | 4 | 5 | 1 | 1 | Todd explores production AI pitfalls and the overhyped agent label. Parag breaks down the unresolved bottleneck of customer evaluation contracts and defines true agency as open-ended tool use. | |
| Parallel's Core Capabilities and the Essential Enterprise Agent Stack | 4 | 5 | 1 | 1 | Todd asks for a breakdown of Parallel's differentiators and the enterprise agent stack. Parag breaks down the top three essentials: access to internal enterprise data, code sandboxes, and the live open web. | |
| Economic Alignment, Differential Pricing, and Protecting the Open Web | 5 | 6 | 1 | 1 | Todd probes how content creators will monetize proprietary data when agents scrape the web. Parag presents a market mechanics solution modeled on differential ad pricing to cross-subsidize open web access. |
Statements from this episode (27)
Agrawal: AIs will use the web 1,000x to 1,000,000x more than humans
“We believe that AIs will use the web a thousand X, a million X, more than humans ever have. And as a result, The web will need to transform.”
Agrawal: Parallel has the only APIs reaching superhuman performance on research benchmarks
“In fact, like the products we've shipped recently are the only APIs that hit human and even higher than human level performance on a variety of deep research benchmarks.”
Agrawal: Parallel is fully in-person five days a week
“Twitter was famously remote friendly. Parallel, we're all in person, in one office, five days a week.”
Agrawal: Building AI products shifts focus from user needs to stochastic feasibility
“People need everything with AI. So that's almost obvious. And now all you really have to zoom into what works and what can work. And where is it going? And you have to look sort of a step ahead and build towards that. And it's a very different art. Like, every…”
Agrawal: Parallel deliberately built a slow API designed for AI agents
“One thing we did early on was we took the counter approach of saying our customer is an AI. And one of the things that led us down was we are going to design an API that is excruciatingly slow.”
Agrawal: CEOs must shape the company around themselves, not adapt
“One of the big shifts I went through taking on the CEO was I decided that the most productive way to operate would be to shape the company around me, take on the founder role for the company, because I think there is no other way. If you try to fit yourself to…”
Agrawal planned major Twitter headcount and roadmap changes within one month
“So one of the things I did as soon as I took the role is I think we changed the structure of the company. We changed the leadership team substantially. Within a month, I was about to meaningfully change how many people were at Twitter and what our roadmap and …”
Agrawal: Twitter planned 20-25% headcount cut just before Musk acquisition
“Like one weird anecdote that I think out there, but not understood is that the three days after we signed the deal to sell the company, which was a Monday, Thursday was a earnings call. And there was a whole plan to really Like trim the company down by about 2…”
Agrawal: X's Community Notes feature is living up to transparent ideals
“And I think as we've seen the team continue to work on like what we used to call birdwatch and now it's called community notes. I think it's great. It's loving up to the top ideals of being transparent, being open and sort of enabling people to contribute to T…”
Agrawal: Parallel aims to ensure AI agents have permissionless web access
“I think that's the same ethos we're taking at parallel, which is like, I think we want the web to be open, and we want it to be permissionless, and we want it to be a free market, and we want everyone to have access to everything, including AIs. We want AIs to…”
Jackson: Founders rarely practice picking ideas, while VCs train it constantly
“As an engineer, as a founder, you, you've developed these muscles on how to build, and you are very good at building. But what you only do once or twice is pick the thing to work on. And VCs, you know, one of the, I think the luxuries that we have VCs, we get …”
Agrawal: AI improves at repetitive tasks while human team performance declines
“And then we realized that AI performance, when you have repetitive, well-specified work, Improves. Human performance, when you have to train 50 people to do repetitive work, declines.”
Agrawal: Parallel serves millions of daily requests ranging from $0.005 to $10
“We now serve, like, millions and millions of requests a day, and it's our cheapest, cheapest requests on our system cost under half a cent. The most expensive requests to our system today can cost 10 dollars.”
Agrawal: AI evaluations must use comparative analysis because ground truth is flawed
“Turns out ground truth is never ground truth, right? So one of the things we sort of totally changed in our worldview was ground truth is never ground truth. And that we actually have to do comparative analysis of quality and really reframe every customer conv…”
Agrawal: Hyperscalers and AI labs move too slowly compared to startups
“I think like hyperscalers
Can't move fast enough to capture the opportunity.
Can't build fast enough.
Right?
The labs are amazing.
They move really, really fast.
But they're also slow relative to how fast we can move.”
Agrawal: Patrick Collison advised that a startup's first API design is always wrong
“One of the pieces of advice I got as I was starting this company, I spoke with Patrick Collins at Stripe. On building an API business, he said, like, your first API design will be wrong. Just know that.”
Agrawal: Choose investors whose disagreements keep you awake reconciling why
“And I had one very simple rubric. People, it's for me just the people I would spend time with. And if they said something to me, and I dramatically disagreed with them, will I feel sleepless that night to try to reconcile why? And people who I feel that way wi…”
Agrawal: Founders should only raise capital that increases binary success odds
“To me, building the company is binary. Either we are successful or we're not. There's just like nothing in the middle, right? So I asked myself the question, I should raise the amount where I can argue up till that amount that I have a reason for believing tha…”
Agrawal: Betting on high-upside hiring requires embracing decisive firing
“The second thing that's really important, if you embrace this sort of bet, take risk on hiring, is you have to be somewhat like, the system only works if you also embrace being decent at firing.”
Agrawal: Parallel operates with no formal roles, teams, or management hierarchy
“We don't talk about roles. We don't have teams. We don't have hierarchy. We're very chaotic.”
Agrawal: Vibe coding without understanding creates bugs and cleanup work for teams
“I think it's a learned skill to be productive with AI in a real production system. And I've personally, I know my team sees this like vibe coded stuff without deep, full understanding of what I was doing. And my team has had to clean up behind me.”
Agrawal: Parallel provides source attribution and confidence scoring on outputs
“The second thing we do really well is really good attribution down to specific sources from which web research was done in a very fine grained granular way. And that allows us to build a differentiator, which is confidence scoring for all of our output element…”
Agrawal: Enterprise agents fundamentally require web, internal data, and code sandboxes
“It must have access to all kinds of internal tools and data in that enterprise. It must have access to the web. It perhaps must have access to, because they're so good at coding, to a sandbox where it can write and run code. Right? And you can come up with a f…”
Agrawal: Web faces existential risk of closing up without new economic incentives
“I think there is a existential risk on the web that the web might close up more and more unless we figure out and solve problems in a way that incentivizes This data to remain open, and it must be optimal for most people for putting their content, their ideas,…”
Agrawal: Parallel tracks source attribution to calculate publisher compensation
“Part of the reason we do so much attribution down to sources is to be able to do that math. To be able to understand feasibility. We really want to create open solutions where everyone has incentive to remain open”