• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    AI

    Meta reportedly plans personal agents for everyday tasks with subsidies

    A user’s summary of a Zuckerberg interview highlights Meta’s existing reach and cash flows as a way to fund up to 100 million free tokens a week plus a virtual computer.

    Nathan BenaichNB
    👩‍💻 Paige Bailey👩‍💻P
    signüllSI
    13 Sources, ,

    TLDR

    According to a user’s summary of a Zuckerberg interview, Meta’s personal-agent strategy targets recurring jobs such as ordering ingredients, building software and managing ads. The account describes connecting social context with email, messages, health and business tools, alongside limited permissions and agents that check activity. It says a confidential virtual machine is being developed to keep its contents hidden even from Meta. The summary also describes monetization through subscriptions and small transaction fees, potentially paid by businesses.

    Combined views

    501.7K

    13 Sources, first seen 26d ago

    Combined views

    501.7K

    13 Sources, first seen 26d ago

    2.8K likes
    26d ago
    first seen 26d ago
    2.8K likes
    178 comments
    569 saves
    110 reposts
    178 comments
    569 saves
    110 reposts

    Sentiment

    Positive40%60%Negative

    Based on 25 sentiment-bearing replies from 25 accounts across 2 conversations.

    Sentiment

    Positive40%60%Negative

    Based on 25 sentiment-bearing replies from 25 accounts across 2 conversations.

    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    13 Sources

    signüll@signulllalso it’s unclear to me that even google can offer up this type of allocation to users given that they have to allocate compute to search / ads / workspace users etc whereas meta has no other crazy use endpoint for ai other than muse + ads. i think meta has a unique position here to really take over consumer agents, esp when integrated with whatsapp & maybe even imessage too? plus the have a desktop / web surface already built too.26d
    Shanu Mathew@ShanuMathew93Zuckerberg lays out Meta’s product strategy for personal agents pretty clearly in this interview w/ @alexeheath @finkd. Would hate to compete with him - super capable and focused. >Distribution + subsidy: billions of existing users, hundreds of millions of small businesses, and cash flows to fund up to 100M free tokens/week plus a virtual computer. Expensive for independent agent companies to compete with. >Personal context + protection: connect social w/ email, messages, health and business tools, with limited permissions, secure credentials and sentinel agents checking activity. A confidential VM is being developed so even Meta can’t see its contents. >Specific, recurring tasks: order ingredients, secure permits, review training footage, build software and manage ads. Remember what worked, suggest improvements and eventually interact with other people’s agents. >Models built for those jobs: coding to execute tasks and discretion to use personal information without unnecessarily sharing it. He describes Watermelon as a significantly more advanced, larger pre-train than Avocado, with further post-training gains to come. >Compute already supporting the next generation: Zuckerberg says the gigawatt-scale Prometheus cluster in Ohio is online and training models beyond Watermelon. He doesn’t specify how much of that capacity is energized or utilized. >Monetization through subscriptions and small transaction fees, potentially paid by businesses. The aim is for agents to make or save users enough money to support the service. Meta is spending to own the models and infrastructure while using its existing business to fund distribution. Zuckerberg is unusually explicit here about the products and economics he wants that investment to enable.24d
    Jeff Weinstein@jeff_weinstein.@google still has a massive opportunity to make a great consumer agent given default trust of gmail—the de facto data warehouse of personal context—and their compute prowess.23d
    Jared Palmer@jaredpalmer@jeff_weinstein @Google Gmail has so much potential but the risk of cataclysmically destroying email with slop also super high. Reminds me of when Google decided they wouldn’t ship a maps feature to optimize scenic routes so as not destroy foot traffic to less developed areas.23d
    Alex Heath@alexeheathRT @kidehen: "If everyone has a strong personal agent, more people can start small businesses without raising a lot of capital. He expects…23d
    David Lieb@dfliebRT @jeff_weinstein: .@google still has a massive opportunity to make a great consumer agent given default trust of gmail—the de facto data…23d
    Sar Haribhakti@sarthakghThe baseline for table stakes in personal agents landscape is going up really fast lately No marginal feature will be a differentiator for too long21d
    Chubby♨️@kimmonismusAs I’ve said before, Meta has a moat against the competition: distribution. Its models and agents don’t need to solve Millennium Prize problems to be successful; they just need to be good enough for 99% of everyday tasks and problems while offering good accessibility. Most users aren't interested in DeepSWE benchmarks or the like; instead, they want their AI to solve everyday problems. Furthermore, Meta has its own ecosystem into which its AI and agents can be deeply integrated. They are on a very good way.20d
    Nathan Benaich@nathanbenaichwhy can't gemini within gmail move files/take actions with files into google drive? still bananas to me how poorly connected google products are20d
    👩‍💻 Paige Bailey@DynamicWebPaige@nathanbenaich Tagging in @ChanduThota (who heads up Google Workspace and its AI features), thank you for sharing the feedback!20d

    13 Sources

    signüll@signulllalso it’s unclear to me that even google can offer up this type of allocation to users given that they have to allocate compute to search / ads / workspace users etc whereas meta has no other crazy use endpoint for ai other than muse + ads. i think meta has a unique position here to really take over consumer agents, esp when integrated with whatsapp & maybe even imessage too? plus the have a desktop / web surface already built too.26d
    Shanu Mathew@ShanuMathew93Zuckerberg lays out Meta’s product strategy for personal agents pretty clearly in this interview w/ @alexeheath @finkd. Would hate to compete with him - super capable and focused. >Distribution + subsidy: billions of existing users, hundreds of millions of small businesses, and cash flows to fund up to 100M free tokens/week plus a virtual computer. Expensive for independent agent companies to compete with. >Personal context + protection: connect social w/ email, messages, health and business tools, with limited permissions, secure credentials and sentinel agents checking activity. A confidential VM is being developed so even Meta can’t see its contents. >Specific, recurring tasks: order ingredients, secure permits, review training footage, build software and manage ads. Remember what worked, suggest improvements and eventually interact with other people’s agents. >Models built for those jobs: coding to execute tasks and discretion to use personal information without unnecessarily sharing it. He describes Watermelon as a significantly more advanced, larger pre-train than Avocado, with further post-training gains to come. >Compute already supporting the next generation: Zuckerberg says the gigawatt-scale Prometheus cluster in Ohio is online and training models beyond Watermelon. He doesn’t specify how much of that capacity is energized or utilized. >Monetization through subscriptions and small transaction fees, potentially paid by businesses. The aim is for agents to make or save users enough money to support the service. Meta is spending to own the models and infrastructure while using its existing business to fund distribution. Zuckerberg is unusually explicit here about the products and economics he wants that investment to enable.24d
    Jeff Weinstein@jeff_weinstein.@google still has a massive opportunity to make a great consumer agent given default trust of gmail—the de facto data warehouse of personal context—and their compute prowess.23d
    Jared Palmer@jaredpalmer@jeff_weinstein @Google Gmail has so much potential but the risk of cataclysmically destroying email with slop also super high. Reminds me of when Google decided they wouldn’t ship a maps feature to optimize scenic routes so as not destroy foot traffic to less developed areas.23d
    Alex Heath@alexeheathRT @kidehen: "If everyone has a strong personal agent, more people can start small businesses without raising a lot of capital. He expects…23d
    David Lieb@dfliebRT @jeff_weinstein: .@google still has a massive opportunity to make a great consumer agent given default trust of gmail—the de facto data…23d
    Sar Haribhakti@sarthakghThe baseline for table stakes in personal agents landscape is going up really fast lately No marginal feature will be a differentiator for too long21d
    Chubby♨️@kimmonismusAs I’ve said before, Meta has a moat against the competition: distribution. Its models and agents don’t need to solve Millennium Prize problems to be successful; they just need to be good enough for 99% of everyday tasks and problems while offering good accessibility. Most users aren't interested in DeepSWE benchmarks or the like; instead, they want their AI to solve everyday problems. Furthermore, Meta has its own ecosystem into which its AI and agents can be deeply integrated. They are on a very good way.20d
    Nathan Benaich@nathanbenaichwhy can't gemini within gmail move files/take actions with files into google drive? still bananas to me how poorly connected google products are20d
    👩‍💻 Paige Bailey@DynamicWebPaige@nathanbenaich Tagging in @ChanduThota (who heads up Google Workspace and its AI features), thank you for sharing the feedback!20d