BREAKING: Ex-Twitter CEO’s New $2B Web for Agents

1. Key Themes

Theme: Agent-driven web usage is in the earliest innings, with 5,000–10,000x growth ahead

Agents currently use the web at a fraction of human levels

Agrawal's original deck projected 1,000x more usage than humans, and he now treats that as the low end. His own heavy agent usage is the proof point, while the broader market is far behind.

"My best estimation right now of where we are is agents are using the web 10 to 20% of humans. Which is to say if my 1,000x number is right, we have from here at least 5,000x or 10,000x growth left."

Supporting evidence: "An average person runs roughly 10 Google searches a day. The agents Agrawal personally tracks run several thousand per day, putting his own usage near 1,000x."

Coding agents show what the adoption curve looks like

Vercel's data gives a concrete preview of how fast agent traffic can overtake human traffic.

"We started out the year with 3% of deployments on Vercel were coming from coding agents. Now, 60% are agentic." - Guillermo Rauch, Founder and CEO, Vercel

Cognition shows the same pattern internally: "Total code shipped inside Cognition grew 7x in 6 months, and customer usage of Devin grew 11x to 12x over the same period."

Theme: Search is becoming a distinct, GPU-intensive infrastructure layer

Search must be re-architected for GPUs, with a fixed ratio to inference spend

Traditional search was CPU-, memory-, and disk-bound. Agent-scale query volume forces a rebuild around distilled models and GPU-based indexes. Agrawal frames his market size as a ratio of inference compute.

"Historically, search was done on CPUs. Memory and CPU bound. Now you get to architect and re-architect search for a price performance band that the world never needed until today."

"For every 10 GPUs the world spends on model inference, we'll spend 1 to 2 on web search. That's the business I'm in."

Agent search is a three-way squeeze on quality, cost, and latency

Early customers like Harvey and Notion pushed quality. Voice agents then collapsed the latency budget from human-search norms to a fraction of it.

"Now you gotta be able to search the web in 150 or 200 milliseconds."

"So now you have to push yourself on the quality dimension, on the cost dimension, and the latency dimension, and you gotta crush all three."

Better search lowers total token cost, which sells to enterprises

Parallel is selling search alongside tokens as enterprises scale model spend.

"Because our search is so high signal and token efficient, the models do half the number of searches, use up far fewer tokens, end-to-end are cheaper, faster."

Evidence of demand: headcount grew "from about 30 people at the start of the year to roughly 120," and more than 100,000 developers build on Parallel.

Theme: The web's business model breaks when agents do the browsing

Attention-based monetization stops working

Ads and per-seat subscriptions depend on human clicks and seat counts. Agents remove the attention while leaving the transaction.

"If agents are doing this, the buying action still remains. Every other action breaks. So you can't monetize attention anymore. So ads business model, per seat subscription business models, these things are not durable."

Pay-per-agent-read compensation as the replacement

Parallel is building an AdSense-like system so publishers have an incentive to keep content open to agents rather than block it.

"We built a system which runs a model, not surprising, to compute how much to pay each content owner every time an agent reads their content."

Platforms will field their own agents, creating friction with personal agents

Fred Wilson of USV expects every platform to offer an agent, with consumers wanting agents that represent them. The article notes "Amazon has blocked Muse from accessing its site."

"We do believe that people will want their own agents that represent them as opposed to my Amazon, only having my Amazon agent." - Fred Wilson, Union Square Ventures

Theme: The web shifts from pull to push, and trust/security becomes the bottleneck

Proactive agents become viable as the cost of intelligence falls

Always-on consumer agents are still mostly reactive. Earlier proactive attempts (Google Alerts, Google Now) failed, but cheaper intelligence changes the economics.

"I've been out there saying the web will go from pull to push."

Compute prioritization becomes a core problem, since the web is roughly a trillion pages, mostly junk and increasingly flooded by agent-generated SEO content.

"No matter how cheap you make compute, we have not enough compute to compute everything."

Agent identity and security require layered defense

Agrawal splits traffic into crawlers, agents, and malicious bots. Incentives between agents and sites are misaligned, so legitimate agents get blocked. Prompt injection and cyber risk are rising.

"It is almost impossible to take it to zero. You have to contain it because it is a knife fight, and the tactics change."

"This whole world will rely on defense in many, many layers."

Mike Mignano of USV flags a concrete vulnerability: "If you give your agent access to your email, which is where all your password reset emails go, they just automatically will be able to access literally every account you own."

2. Contrarian Perspectives

Ads and per-seat SaaS pricing are "not durable" in an agent-first web

This runs against the assumption that existing monetization models simply carry over. Agrawal's reasoning is that agents preserve the purchase but eliminate the attention that ads and seat-based pricing depend on.

"Every other action breaks. So you can't monetize attention anymore. So ads business model, per seat subscription business models, these things are not durable."

Supporting facts: Fred Wilson "has also described Google's traditional search business as being cannibalized by AI and agents, with Gemini as Google's response."

Bot detection may be solving the wrong problem

The bot-detection industry assumes non-human traffic is a threat to filter out. The article notes Stripe co-founder John Collison asked "whether bot detection is still needed once most web traffic comes from agents." Agrawal's framing supports the tension: detection vendors are growing, yet "consumer agents like Instinct and Muse [are] blocked from sites their users can access," and "if those agents declared themselves today, more sites would block them." The implication is that the core problem is aligning incentives and building agent identity, not better blocking.

Search will not need a Google-scale build-out to serve 1,000x the queries

The naive assumption is that 1,000x query volume requires 1,000x the infrastructure. Agrawal rejects it.

"Google serves an estimated 10B searches a day on infrastructure built over decades. Agrawal does not expect anyone to build 1,000x that footprint to serve 1,000x the queries."

His answer is distillation of ranking intelligence into progressively smaller models running on GPUs, rather than brute-force scaling of legacy architecture.

3. Companies Identified

Parallel Web Systems

• Description: Builds web search and research APIs for AI agents; founded in 2023 by Parag Agrawal.
• Why mentioned: The subject of the interview and the thesis that agents will become the web's primary users.
• Quotes:
  • "Parallel was founded in 2023 and has raised $230M, most recently a $100M Series B at a $2B valuation in April 2026."
  • "More than 100,000 developers build on Parallel, including teams at Harvey, Notion, Replit, & Rogo."

Browserbase

• Description: Supplies cloud browsers that agents use to log in, fill out forms, and complete transactions.
• Why mentioned: Parallel became the default search provider for Browserbase Search API and Agents on the day of recording.
• Quotes:
  • "We now power search in Browserbase, and if you think about every AI agent that people use, what are they doing? They are looking at your internal information, they're looking at web information."

Index Ventures

• Description: Venture investor; host of the recording.
• Why mentioned: Co-led Parallel's $100M Series A and participated in the Series B.
• Quotes:
  • "Index Ventures co-led the $100M Series A in November 2025 and participated in the Series B."

Vercel

• Description: Developer platform for deploying web applications.
• Why mentioned: Case study for the speed of agent adoption in deployments.
• Quotes:
  • "Coding agents triggered 3% of deployments on Vercel at the start of 2026 and now trigger 60%."
  • "Vercel took 10 years to reach its first 1B deployments and has added 1.5B more in the last 10 months."

Cognition (Devin)

• Description: Maker of the Devin AI coding agent.
• Why mentioned: Example of an agent building its own product at scale.
• Quotes:
  • "We almost exclusively used Devin to build Devin at this point. 95% of the code or something is written by Devin."

Union Square Ventures (USV)

• Description: Venture firm with a view on agents, platforms, and consumer AI.
• Why mentioned: Partners offer perspective on platform agents, consumer AI engagement, and agent security.
• Quotes:
  • "Rebecca Kaden of USV puts consumer AI trial behavior at an all-time high and engagement at an all-time low."

Common Fabric

• Description: USV portfolio company.
• Why mentioned: Addresses the agent-permissions security gap.
• Quotes:
  • "USV portfolio company Common Fabric builds a sandbox for agent permissions."

Stripe

• Description: Payments company co-founded by John Collison.
• Why mentioned: Collison's question at Browserbase Navigate challenged whether bot detection remains necessary.
• Quotes:
  • "At Browserbase Navigate, Stripe co-founder John Collison asked whether bot detection is still needed once most web traffic comes from agents."

Harvey

• Description: AI platform for legal work.
• Why mentioned: Early Parallel customer pushing search quality for high-stakes work.
• Quotes:
  • "Early customers such as Harvey and Notion were building agents for high-stakes work, which pushed Parallel's systems on quality."

Notion

• Description: Productivity software company.
• Why mentioned: Parallel customer that pushed quality requirements.
• Quotes:
  • "More than 100,000 developers build on Parallel, including teams at Harvey, Notion, Replit, & Rogo."

Consumer agents (Instinct, Muse, Dot)

• Description: Always-on consumer AI agents.
• Why mentioned: Examples of reactive-but-evolving consumer agents, and of agents being blocked by sites.
• Quotes:
  • "Consumer agents including Instinct, Muse, and Dot are always on but still mostly reactive."

Google

• Description: Incumbent search provider.
• Why mentioned: Baseline for search scale and a business under pressure from agents.
• Quotes:
  • "Google serves an estimated 10B searches a day on infrastructure built over decades."

Other Parallel customers (Lovable, Granola, Replit, LangChain, Profound, Modal, Rogo)

• Description: AI and developer-tool companies using Parallel's search infrastructure.
• Why mentioned: Evidence of customer breadth in agent-native software.
• Quotes:
  • "building AI-native search infrastructure for customers like: Harvey, Lovable, Granola, Replit, LangChain, Profound, Modal, Notion & Browserbase."

4. People Identified

Parag Agrawal

• Description: Founder and CEO of Parallel Web Systems; former Twitter engineer, CTO, and CEO.
• Why mentioned: Interview subject and architect of the agent-web thesis.
• Quotes:
  • "Parag Agrawal is the Founder and CEO of Parallel Web Systems... He spent 11 years at Twitter as an engineer, CTO, and CEO."
  • "I've been out there saying the web will go from pull to push."

Guillermo Rauch

• Description: Founder and CEO of Vercel.
• Why mentioned: Provided data on the share of deployments triggered by coding agents.
• Quotes:
  • "We started out the year with 3% of deployments on Vercel were coming from coding agents. Now, 60% are agentic."

Scott Wu

• Description: Co-Founder and CEO of Cognition.
• Why mentioned: Evidence of agents writing the majority of a company's own code.
• Quotes:
  • "We almost exclusively used Devin to build Devin at this point. 95% of the code or something is written by Devin."

Fred Wilson

• Description: Union Square Ventures investor.
• Why mentioned: Views on platform agents versus personal agents and Google's search disruption.
• Quotes:
  • "We do believe that people will want their own agents that represent them as opposed to my Amazon, only having my Amazon agent."

Rebecca Kaden

• Description: Union Square Ventures partner.
• Why mentioned: Observation on consumer AI trial versus engagement.
• Quotes:
  • "Rebecca Kaden of USV puts consumer AI trial behavior at an all-time high and engagement at an all-time low."

Mike Mignano

• Description: Union Square Ventures partner.
• Why mentioned: Flagged email access as a major security gap for consumer agents.
• Quotes:
  • "If you give your agent access to your email, which is where all your password reset emails go, they just automatically will be able to access literally every account you own."

John Collison

• Description: Co-founder of Stripe.
• Why mentioned: Questioned the long-term need for bot detection.
• Quotes:
  • "John Collison asked whether bot detection is still needed once most web traffic comes from agents."

Molly O'Shea

• Description: Author and host of the Sourcery newsletter and interview.
• Why mentioned: Interviewer and source of the piece.
• Quotes:
  • "Recorded at Index Ventures in San Francisco, we discuss the massive GPU demand behind AI search..."

5. Operating Insights

Design your product and onboarding for agent users, not just humans

Parallel built its website for agent readers from launch and structured onboarding around a developer working with a coding agent. Vercel does something similar by serving agents a text-based version of its site.

"Parallel has designed its own website for agent readers since launch, with a toggle between human and machine versions. Its onboarding assumes a developer working with a coding agent, providing prompts that let the agent set up Parallel while the human authorizes access, adds payment, and accepts terms."

Treat GPU procurement as a specification problem and a standing capability

Parallel has run repeated sourcing fire drills and has never turned away a customer for lack of capacity. Each request specifies count, chip type, timing, location, and duration, and location is often the binding constraint. Speed of relationship-based sourcing matters because the market is inefficient.

"Sometimes we need GPUs within 5 milliseconds of where our large web index is sitting."

"You're figuring out availability on iMessage, and then the finance person calls the finance person, and the paperwork gets done."

Build go-to-market around three distinct buyer entry points

Parallel built GTM from zero over 3 to 6 months and identified where customers come from, which makes the sales motion repeatable: (1) teams whose bundled model search limits output quality (legal, finance, science), (2) teams using multiple models that need a model-agnostic search layer, and (3) teams at scale seeking lower end-to-end cost.

"Customers arrive at 3 points. Some find that the search bundled with their model limits output quality... Others adopt multiple proprietary or open weight models and need search that works across all of them. A third group runs scaled workloads and wants to reduce end-to-end cost."

6. Overlooked Insights

The latency tolerance window for voice agents closed in about three months

Users briefly accepted voice agents that paused to look something up, then expectations reset quickly. This suggests user tolerance for AI product shortcomings is short-lived, and infrastructure vendors need to stay ahead of the tolerance curve rather than react to it.

"Users briefly accepted voice agents that paused to look something up, and that tolerance lasted about 3 months."

Web search is already table stakes for LLM usefulness

Agrawal's estimate that removing search would break 90% of queries implies search is a core dependency of every agent product rather than an optional feature, which underpins the durability of the infrastructure layer he is building. The article notes ChatGPT launched without it, for roughly its first 8 to 9 months.

"He estimates that turning web search off today would leave 90% of queries unsatisfactory."