Korva is a scroll of the internet’s best ideas.
Google Cloud adds Parallel search
The Google Cloud deal
Parallel Web Systems builds search for AI agents and was founded by Parag Agrawal, formerly the CEO of Twitter. Under the arrangement, developers who build agents or chat apps on Google Cloud, or run other LLM inference there, can attach web search to Gemini or other models available on the platform. Their options are Google Search or Parallel Search.
Built into the product
The integration is built into the product. Technical, training, product, and commercial teams from both companies worked on it so that Gemini models used with Parallel return strong results.
A faster product
Parallel also shipped a new product last week that returns agentic web search in 200 milliseconds. Until then, the company gave itself three seconds to throw compute at a query at read time.
How does grounding reduce made-up answers?
Grounding gives a model real documents to draw on while it writes an answer, instead of relying only on what it memorized. Researchers at Facebook AI, led by Patrick Lewis, published the retrieval-augmented generation method in 2020. The model retrieves relevant passages first and then writes its response using them. This lets answers cite current sources that were never part of training.
How does a model call web search?
Through a feature called function calling, a model can output a structured request instead of ordinary text. The software around the model reads that request, runs the search, and sends the results back. The model then continues its answer with those results in view. Developers define which tools the model may use, so search is one option among many.
Why does speed matter for agent search?
An agent that answers a question often runs several searches one after another. Each search must finish before the next one can be planned, so their delays add together. A faster search shortens the whole chain, not just a single step. In user interfaces, response times near a tenth of a second are widely treated as feeling instant.