When Elon Musk took over Twitter (now X) in 2022, Parag Agrawal abruptly stepped down from his place as CEO and left the corporate to pursue one thing new. Three years later, Agrawal is again with a brand new enterprise that’s unsurprisingly centered on AI.
Known as Parallel Net Programs, Agrawal’s new startup intends to construct instruments and infrastructure that allows AI programs to work together with the open net in real-time. This web-scale infrastructure has been constructed from the bottom up particularly for AI brokers to fetch, confirm, and organise the knowledge they current.
“AIs will use the online excess of people ever have. As a lot as we’d anthropomorphise AIs, they function in a different way from people on the net. […] At Parallel, we’re constructing for the online’s second consumer. We’re creating programs and infrastructure for AIs to make use of the online successfully for finishing complicated duties,” the corporate mentioned in its announcement weblog submit.
“Brokers are going to be the first clients of the online going ahead. They may use the online much more than people ever have,” Agrawal, the founder and CEO of Parallel, mentioned.
Parallel’s deep analysis APIs
Parallel’s suite of merchandise embrace low-level search instruments and deep analysis APIs that deploy AI brokers to finish hours of human work on the net in minutes.
The corporate claimed that its deep analysis APIs can outperform main AI fashions, together with OpenAI’s GPT-5, on benchmarks testing complicated net search and analysis duties.
When evaluated on OpenAI’s BrowseComp benchmark for accuracy in multi-hop reasoning and net navigation, Parallel’s deep analysis APIs achieved 58 per cent accuracy whereas GPT-5 recorded 41 per cent accuracy. Human individuals working with two-hour deadlines achieved 25 per cent accuracy.
Story continues beneath this advert
On DeepResearch Bench, a benchmark take a look at to measure the standard, depth, and rigor of long-form AI-generated analysis stories on complicated subjects throughout domains, Parallel secured 82 per cent win charge towards a reference set, in comparison with a 66 per cent win charge by GPT-5.
Parallel mentioned it is ready to obtain these outcomes as a result of its web-scale infrastructure has been constructed particularly for AI brokers, with every layer of its stack, from crawl and index to rating, engineered for “how machines devour info, not how people browse.”
A number of startups, public enterprises, and AI firms are already utilizing Parallel’s deep analysis APIs to execute “hundreds of thousands of analysis duties each day,” as per the corporate. “Builders are constructing AI gross sales brokers that analysis leads, coding brokers that synthesize context from docs, and funding instruments that discover alpha in good components of the online and SEC filings. Insurance coverage firms are automating claims with web-sourced verification,” it mentioned.
What can we find out about Parallel?
Parag Agrawal studied at Indian Institute of Know-how (IIT) Bombay earlier than incomes his PhD at Stanford College in the USA. He labored at Twitter (now X) for practically ten years.
Story continues beneath this advert
Parallel Net Programs is Agrawal’s first main undertaking since his tenure at Twitter abruptly resulted in 2022 following Elon Musk’s $44 billion takeover of the social media platform.
The startup has raised roughly $30 million in funding thus far. It’s backed by outstanding Silicon Valley enterprise capitalists (VCs), together with Khosla Ventures, Index Ventures, and First Spherical Capital,
The corporate mentioned that its founding crew consists of engineers who “beforehand constructed the infrastructure and markets that powered the human net” at Twitter, Google, Stripe, Airbnb, Chime, Waymo, and Kitty Hawk.
Within the days to return, Parallel mentioned it’s going to look to develop long-horizon AI brokers that may full the work of total groups in hours, steady monitoring programs that monitor the online for alerts, event-driven architectures, SQL-style programmable queries over the online, and extra.