Oct 24, 2026, 9:02 AM ETInterview

AI Agents Could Drive 10,000x More Web Activity | Former Twitter CEO

Parallel Web Systems closed a $100 million Series B funding round last month, achieving a $2 billion valuation.
The company is now three years old and has grown its team from approximately 30 employees at the start of the year to 120 currently.
Total reported funding to date is $230 million, though the specific breakdown between Series A and B was not fully detailed.

Founder (former Twitter CTO/CEO) predicts agents will use the web 1,000x more than humans; current personal data suggests the ratio is already approaching 1,000x (thousands of agent searches vs. <10 human searches daily).
Current estimates place agent traffic at 10–20% of human web traffic, implying 10x to 20x growth remains before parity is reached.
The core thesis involves a "parallel web" where agents read and write to the web 1,000x more than humans, fundamentally shifting infrastructure needs.
Expected growth rate for Parallel is tied to the year-over-year growth of GPU inference, with a short-term benchmark goal of 10x growth.

Parallel does not aggregate GPUs but rents/buys them directly from hyperscalers and neoclouds to support high-volume search inference.
Search architecture has shifted from CPU-bound to GPU-based, allowing for optimized models that distill intelligence into smaller, cheaper units for ranking.
Latency requirements for voice agents have tightened to 150–200 milliseconds, significantly faster than the ~700ms standard for human web search.
The company employs a "human-versus-machine" toggle on its website, transforming unstructured human-readable content into agent-friendly structures automatically.
Parallel optimizes token efficiency, enabling models to perform tasks with half the number of searches and significantly fewer tokens compared to bundled solutions.

Traditional web monetization (ads, subscriptions based on human clicks) is becoming unsustainable as agents bypass human attention spans.
Parallel is building a new "AdSense-like" system that pays content owners via auto-differential pricing every time an agent reads their content.
The goal is to solve "incentive alignment" so content owners are motivated to keep data open to agents rather than blocking them.
There is a strategic shift from "SEO" (Search Engine Optimization) to "Agent Experience Optimization" (AEO), where sites are built specifically for agent consumption.

The company distinguishes itself by enabling "model independence," allowing clients to switch between proprietary or open-weight models without changing their search layer.
Key customer segments include high-barrier industries (legal, finance, science) where trust and quality of agent output are critical.
Major competitors in the space include Instinct, Muse (Jolly), and OpenAI's "Dots," which the founder praises for their rapid user growth and natural onboarding.
The product strategy assumes the user is a developer with a coding agent, designing documentation and onboarding flows for agents to execute tasks automatically.

The founder argues that "bot detection" is becoming obsolete; the focus should shift to "agent identity" where good agents declare themselves and inherit human privileges.
Security is viewed as a multi-layered defense problem requiring obsession at the web layer, model layer, and application layer.
Cybersecurity risks include "injection attacks" where malicious content flows into an agent's context, though models have made strides in vulnerability reduction.
A "weakest link" dynamic persists where attackers target less secure implementations, requiring constant adaptation.

The founder attributes his drive and ambition to his wife, Benita (a physician and investor at a16z), who acts as a high-expectation peer and standard-setter.
He rejects the concept of "pressure," viewing it as meeting external expectations, and instead champions "insecurity" as a productive state that drives learning and humility.
The company's hiring philosophy focuses on "outliers," operating on the principle that high talent density is the only way to attract more outliers.

The web is transitioning from a "pull" model (humans searching) to a "push" model where proactive agents monitor the web for actionable changes.
Future compute allocation must prioritize high-signal changes over the noise of agent-generated content, which is increasing the web's rate of change.
The founder predicts agents will eventually become hyper-proactive, anticipating user needs and acting before being prompted.
Long-term infrastructure needs will involve sovereign participation in data centers to manage compute at the scale of a trillion web pages.