Ex-Twitter CEO's Parallel raises $30M to build an internet for AI

The modern web was built for human eyeballs, blue links, and ad clicks. But what happens when the primary users browsing the internet are no longer people, but AI agents devouring entire databases in seconds?

Former Twitter CEO Parag Agrawal is tackling that exact problem. His stealth startup, Parallel Web Systems, has raised $30 million from heavyweight investors including Khosla Ventures, First Round Capital, and Index Ventures.

Founded in 2023 and operating with a lean 25-person team in Palo Alto, Parallel is building what it calls a new "Programmatic Web" specifically designed for AI systems.

Agrawal announced on LinkedIn that the company just launched its Deep Research API. He claims the tool outperformed both humans and leading frontier models, including OpenAI's GPT-5, across two of the toughest research benchmarks.

Turning data pulling into automated pushing

Agrawal’s core thesis is straightforward: the open web is facing an existential crisis because business models built around human attention cannot serve machines.

While people browse casually, AIs interact at massive scale. Today, that collision leads to paywalls, gated APIs, and private silos that break agent workflows.

Parallel flips the standard browsing dynamic. Instead of an AI agent writing brittle scripts to scrape raw HTML, the agent simply specifies what output it needs. Parallel’s infrastructure handles the complex web discovery, computation, and reasoning behind the scenes, delivering structured intelligence with verifiable provenance.

According to Agrawal, Parallel is already processing millions of research tasks every day. Fast-growing AI startups and autonomous coding agents are actively relying on its search layer to pull fresh technical docs and debug runtime errors.

Why it matters

We think Agrawal is targeting one of the messiest, most frustrating pain points in modern software: keeping AI agents grounded with real-time web knowledge without constantly breaking against bot detection and anti-scraping walls.

For developers building autonomous agents, having an off-the-shelf research API that handles web aggregation means you can skip building complex scraping pipelines. If you are building research assistants or developer tools, this is definitely worth testing.

Strategically, Parallel is taking a smart, Switzerland-style infrastructure approach. While tech giants like OpenAI and Google package search directly into their proprietary models, Parallel wants to be the neutral plumbing layer powering everyone else.

That said, the biggest question mark remains sustainability. If AI agents consume the web without feeding ad revenue or human attention back to original creators, publishers will only build higher walls. Parallel's vision of an open, value-based market will only succeed if content owners actually get paid.