Our next public workshop, AI Tool Mastery for Work Productivity, starts on Tuesday, 24 November 2026 (Kota Kinabalu, Sabah). Click to view and register.
    Starts In
    Back to blog
    AI News

    Box's AI Document Search Can Now Answer Four Times Faster

    By Edison Chua 7 November 2026 4 min read

    Box just built AI into document search so agents answer requests four times faster and far more accurately.

    Box's AI Document Search Can Now Answer Four Times Faster

    Image: TechCrunch

    Quick Summary

    • Box announced Content Retrieval on 5 November 2026, built on Nvidia's Nemotron AI models.
    • Box expects AI agents to answer document questions about four times faster, with 20 percent fewer errors and 90 percent fewer tokens used.
    • The feature rolls out to Enterprise Advanced customers starting early next year and respects a company's existing file permissions.
    • Also today: Meta's Muse agent gained Spaces for building trackers and planners, Palo Alto Networks is reportedly paying 500 million dollars for AI agent security startup Clutch, and Ramp data shows AI spending growth slowing outside big tech.

    Box announced a new AI document search feature called Content Retrieval on 5 November 2026, built on Nvidia's Nemotron AI models. Instead of making an AI agent read an entire file, the system pulls back only the most relevant chunks, so Box says agents answer about four times faster, with 20 percent fewer errors and 90 percent fewer tokens used.

    What is Box's AI document search for business?

    Ordinary keyword search only matches the exact words you type, so it misses a document that describes the same idea in different wording. Nemotron is Nvidia's embedding system, which topped the RTEB retrieval benchmark when Nvidia released it in July 2026. Box built Nemotron into its existing document store so an AI agent gets a map of the ideas inside a file, not just its keywords. A search for a termination clause can then also surface a document that talks about ending the agreement, because the system understands the concept behind the words.

    This dramatically boosts both the relevance of our search technology for agents.from Aaron Levie, CEO, Box (TechCrunch, 5 November 2026)

    What can it actually do?

    • Sends an agent only the document chunks that answer the question, instead of making it read the whole file
    • Understands related ideas even when the wording is different, because Nemotron maps concepts rather than exact keywords
    • Works inside Box's existing folders and permissions, so an agent only ever sees files the employee is already allowed to open
    • Rolls out to Enterprise Advanced customers starting early next year

    How could your business use it?

    Box gives the example of a lawyer searching for a relevant clause or precedent across thousands of contracts, and CEO Aaron Levie described the result as sending back the chunks from the documents that are most relevant to the user's query, rather than having an agent read every file. The same idea applies to any business that keeps its paperwork digital. An accounting firm could ask an agent to pull every invoice tied to one supplier. A clinic could ask it to find the latest version of a consent form. A restaurant chain could ask it to summarise every supplier contract due for renewal this quarter. The real benefit is not the model itself, it is what happens when a team can trust the answer an agent gives back, because it read the right paragraph instead of guessing from a keyword match.

    Also today

    • Meta launched Spaces inside its Muse AI agent, letting users describe an idea in chat and have Muse build a tracker, planner or project board for it (Meta, via TechCrunch, 6 November 2026).
    • Palo Alto Networks is reportedly paying 500 million dollars for Clutch, a startup that controls which company systems an AI agent is allowed to touch (TechCrunch, 5 November 2026).
    • New data from spend management firm Ramp shows most companies' AI token spending barely grew in August and September, a sign that heavy AI spending is still concentrated in a small group of big tech firms (TechCrunch, 6 November 2026).

    What to do this week

    Open whatever system holds your contracts, SOPs or client files this week and time how long it actually takes your team to find one specific answer inside them. If it takes more than a minute, or the wrong document comes up first, that is the exact gap this wave of AI document search tools is built to close, whether you use Box or not. You do not need an enterprise budget to start: ChatGPT and Claude can already search and summarise a folder of your own documents today. That is exactly what we cover hands-on in our AI Tool Mastery for Work Productivity program in Kota Kinabalu, where owners and teams learn to put AI to work on the documents, finance and admin tasks already sitting in their business.

    Common Questions

    What is Box Content Retrieval?
    Box Content Retrieval is a feature Box announced on 5 November 2026 that uses Nvidia's Nemotron AI models to help AI agents search company documents. Instead of reading an entire file, an agent receives only the most relevant chunks, which Box says makes answers faster, cheaper and more accurate for tasks like finding one clause in a long contract.
    How much faster is Box's new AI document search?
    According to Nvidia's VP of generative AI software, Kari Briski, speaking to TechCrunch on 5 November 2026, Box expects the system to answer queries about four times faster, with 20 percent fewer errors and 90 percent fewer tokens used than an agent reading whole documents.
    When can businesses start using this AI document search?
    Box says Content Retrieval is rolling out to Enterprise Advanced customers starting early next year. It works inside Box's existing folders and permissions, so an AI agent only ever sees the files an employee already has access to.

    Source: TechCrunch

    Share
    Edison Chua

    Written by

    Edison Chua

    HRD Corp certified AI trainer and founder of NextGen Training Academy, Kota Kinabalu. Trains Malaysian businesses to put AI to work - hands-on, in plain language.

    Want To Put This To Work In Your Business?

    That's exactly what we teach, hands-on and HRDC-claimable.

    View Programs