Announcing our $100 million Series A at a $740 million valuation to build the web for its second user: AIs. The round was co-led by Kleiner Perkins and Index Ventures, with participation from Spark Capital and support from existing investors Khosla Ventures, First Round Capital, and Terrain. Mamoon Hamid of Kleiner Perkins joins Vinod Khosla, Shardul Shah and Josh Kopelman on the board.
## **The web’s second user: AI**
Two years ago, when we started the company, there weren’t any agents on the web. The prevailing view was that large language models would render the web obsolete. Why would agents need to search when every model had the entirety of the internet in its training data?
We believed the opposite: AIs would use the web far more than humans ever have.
## **Quality is everything**
Today, we offer two types of products, all of which are best-in-class for quality and accuracy of outputs:
- - **Web Tools** built as primitives for search[search](/blog/introducing-parallel-search), extraction[extraction](https://docs.parallel.ai/extract/extract-quickstart), and retrieval for agents to use.
- - **Web Agents **built for structured enrichment[structured enrichment](/blog/parallel-task-api), deep research[deep research](/blog/deep-research-benchmarks), and workflow automation[workflow automation](/blog/case-study-gumloop).
Agents are our users. AI-native builders are our customers.
The most sophisticated builders choose us: Clay, Sourcegraph, Owner, Starbridge, Actively, Genpact, and leading Fortune 100 companies. They've tested alternatives and understand their agents' needs. Whether it's Claygent[Claygent](https://www.clay.com/claygent) powering GTM workflows, Amp[Amp](https://ampcode.com/)'s coding agent solving bugs, helping lawyers find precedent, Starbridge[Starbridge](/blog/case-study-starbridge) discovering government RFPs, or a Fortune 100 insurer underwriting claims. Our customers understand something fundamental: if your agent doesn't have good, fresh, accurate data from the web, all else downstream doesn’t matter.
Parallel drops hallucination rates[drops hallucination rates](/blog/benchmarks-task-api-sealqa) across workflows by organizing and bringing forth the right context from the web. Every decision, by an agent or the human, is downstream of the quality of the data it uses. When every company is racing to deploy AI, competitive advantage comes down to who has better access to accurate information.
## **Search for AIs is very different**
Search[Search](/products/search) for agents differs fundamentally from search for humans. Instead of just keyword queries, AIs can ask declarative queries. Instead of ranking URLs that humans might click, we're identifying the optimal tokens from the web to place in an agent's context window. Instead of being limited to a few hundred milliseconds and a limited amount of compute, we can choose to flexibly allocate computation and time in order to improve outputs.
We’ve built the only web APIs that are optimized to be used as tools within AI agents, and they are used at scale including by our own search agents. Building this product is only possible through innovations across crawling, indexing, retrieval and ranking - each purpose-built for AIs as the primary user.
## **The web must remain open to AIs**
As the web's primary user shifts from humans to AIs, business models built on human attention are challenged. By default, the web trends towards zero-sum: paywalls, private data silos, and gated access. We’ve specifically designed our APIs and systems to not just serve AIs as customers, but also provide content and data owners incentives to continue publishing on the open web, and providing all AI agents broad access.
Our mission is to ensure an open, transparent, and competitive web that can be used by all AIs.
By Parallel
November 12, 2025
