Why Billions of AI Agents Could Beat AGI | The Next Endeavor 2026

Vídeo em Inglês, publicado pelo canal Imagination in Action.

About this content: The summary, key points and topics below are generated automatically from the video transcript. They reflect the content, opinions, claims and predictions presented in the video and do not represent a position, validation or fact-check by IA Dots.

Video summary

The video discusses the concept of artificial general intelligence (AGI) as an investor pitch deck and explores the transition from massive foundation models to a decentralized network of personal and corporate AI agents. It covers challenges such as agent safety, identity, testing, and edge computing, alongside real-world initiatives like India's Digi-Doot project.

Key points

• AGI is described as an investor pitch deck concept rather than a realistically attainable zero-shot reality, given that the world's most valuable data remains confidential.
• Foundation models are compared to car engines, while agents represent the cars, requiring separate safety frameworks, harnesses, and third-party scaffolding.
• The Nanda framework proposes four major challenges for decentralized AI agents: registries, crash test tracks for safety, attestation, and verification.
• India is launching 'Digi-Doot', a project aiming to provide an AI agent for every citizen to assist with civic tasks, healthcare, and pensions using local language support.
• Most agents can run locally on edge devices like PCs, laptops, and phones, reducing the need for massive new data center infrastructure for billions of users.
• The future of AI is framed as moving from a 'factory era' of massive centralized models to a 'bazaar era' where individuals and organizations build unique, interconnected agents.

Topics

AGIAI AgentsDecentralized ArchitectureAgent SafetyDigi-DootEdge ComputingBazaar Era

Video transcript

Transcript from available captions; it may contain errors.

I think everybody knows that this is just a investor pitch deck called AGI. There's no such thing >> Okay, everybody knows AGI is an investor pitch deck [music] thing. I tried to make an agent and before I knew it, the agent was emailing journalists about [music] my little thing that was not ready for prime time. >> So a lot of the work we do at MIT is basically thinking about what does it mean to build a city of agents and not an individual agent. >> [music] >> Let's talk about AGI. >> Yeah, so I I think everybody knows that this is just a investor pitch deck called AGI. There's no such thing >> Okay, everybody knows AGI is an investor pitch deck thing. >> And you know, it you know, it's not possible to create a zero shot AGI because most of the world's knowledge is not going to be available to, you know, any model for a very, very long time. So, you know, if Pfizer has, you know, amazing drug research, they're [snorts] not going to reveal that. Or if a car company has simulation, you know, kind of a dynamic simulation or fluid simulation, they're not going to reveal that because simulation is not even data that they can sell. >> So, like the world's most valuable data, they're not training on. >> Yeah, they just going to do that. >> to AGI if you can't >> There's no zero shot AGI that you can build. The best case scenario is you could buy a product from a company like this and train it locally for your own models, but keep it confidential to your own services. And that's why you could call them agents or local AIs. Uh and those agents, so Pfizer could have an agent for their own research. Uh but that agent can talk to the rest of the world as a service provided by Pfizer but not by OpenAI and Anthropic. >> So, how can Pfizer be confident that this agent is going to do what it is told? >> Yeah, so I mean the analogy here is um the models, especially foundation models, are like an engine of a car. The agents are like a car, but we still have to build safe roads, like the internet of AI agents. So, the company that builds the engine may not be the company that actually builds the car. The agents. So, it's quite likely that, you know, a pharma company will start buying the agent and the frameworks, or sorry, agent harnesses, and the whole [snorts] whole scaffolding around it from a completely third-party player. It doesn't have to be the same companies. >> And perhaps that might be ideal. >> That would be ideal because then they will be focused on not model safety, but agent safety. And those two things are completely different. So, the few of the things we're doing as part of Nanda, which stands for network AI agents in decentralized architecture. I know that's a mouthful, but Nanda is also name of my sister. So, short short for Nanda doesn't mean joy in Sanskrit. And so, for Nanda, we think that we need at least four um major challenges we have to solve. One is that uh you know, like an ICANN or registry. So, you know, the the engagement >> for for the agents, that that every agent is registered. >> And has effectively a driver's license or a passport. >> Right. Uh-huh. >> Second problem we have to solve is uh a notion of a Nanda town, which is um you know, just as when cars are leased, you have to take them to crash test dummies and test tracks, you know, a national highway uh um you know, uh >> NHTSA training and uh safety uh driving test or crash the driving into the wall. >> Exactly. >> No, so >> You need all those things, and finally we need attestation, which is like if somebody claims, "I'm a you know, an agent for diabetics, or I'm an agent for financial advisor." Somebody else has to certify that, and that has to be attested on a minute-to-minute basis. So, if somebody says, "I'm going to give you financial advice." They have to take a test every minute to prove that they're is advisor. >> And so that agent is sort of being constantly verified its authenticity. I tried to build an agent. I want to just pause for a moment and disaster is big word for what it was, but it was an embarrassment. I am a journalist and I have been trying to build an agentic tool for investigative reporters. Well, for anybody to become an investigative reporter using publicly available municipal county and state data. And I use AI to help me make a little marketing plan with it and I've been trying to play around with it and I tried to make an agent and before I knew it the agent was emailing journalists about my little thing that was not ready for prime time. I as a journalist have spent every day for 25 years building my credibility one day at a time. To have an agent go out representing me with some could have been catastrophic. Some of the work that you're doing in India. Can you tell us a bit about that? >> Yeah, I'm glad you mentioned it because I was just in India last week and I came back and >> [snorts] >> Indian government is launching a project called Digi-Doot. Doot means a messenger. And an AI agent for every Indian citizen. 1.5 billion people. >> Every Indian citizen will get an AI their own agent. >> Yes. >> And what will it presumably be able to do on their behalf? >> So in the beginning it might use only for their civic life, you know, kind of their permitting and taxes and driver's license and pension schemes and >> Oh, wait, banking and health care. >> Yes, but only for the government. >> I see. Taxes, paying taxes. >> but then there'll be also, you know, a 70-year-old woman in rural Bihar who doesn't even speak English properly. She can just pick up the phone in her own language and say, "Oh, you know, what are some new pension schemes that are good for me?" And you know, the agent can work for her behind the scenes to do that. But more importantly, one question everybody here would ask like, "Well, for 1.5 billion people, what's the compute cost? You know, where's the AI infra build out? Where's the terawatt of power?" And that's also BS because uh most of the agents have to run on the edge. They can run on your PC, your laptops, even your phones uh at some point. Uh you might run a replica as a backup in the cloud, but you do not need the AI infra build out for billions of agents. >> Um all No, we don't need any new data centers for this? >> We need the data centers for backups, and we might need them for, you know, extremely complex tasks like, I don't know, drug research or climate modeling, you know, new video models and things like that. But for 99% of the population, like students or business owner and so on, you know, very specific models will be good enough. And every once in a while when they do need something that's not running on an agent on their own device, they might make a call to a bigger model that's in the cloud, but majority of the time uh their needs are are are served uh on what's available uh what's available locally. >> In this sort of it's a marketplace for agents. I've I've heard different sort of theories of about this possible model. Um what do you see as the biggest barrier to actually executing this, getting corporations that might not want to be out there getting some random agent from the bazaar? >> Yeah. >> Right? Or or what what's the challenge? >> Yeah, I'm glad you used the word bazaar because how many have read the book The Cathedral and the Bazaar? Um I think right now we are kind of I would say the factory era versus the bazaar era. The factory era is where we assume that certain companies are going to create these massive models and we're going to all use them versus the bazaar era where every one of us is going to create our own agent, and we're going to interact with each other. You know, I might have some expertise on my own agent. You have expertise, you know, Pfizer has their own model, JetBlue has their own model, and we all talk to each other and get things done. And that bazaar era requires a very different preparation. So, if you're a, you know, if you're a if you're Spain or Indonesia or Nigeria or a or a company, don't worry about creating your own models. You know, worry about how are you going to create the ecosystem that sustains the bazaar. You know, how would you create, you know, the identity rails, the payment rails? How would you create the justice system for agents if agents go rogue? What's the agent insurance players? >> Or when when when? >> And and if they if they build all those things, they'll be ready for the bazaar era of AI as opposed to worrying about copying >> buying it off the shelf from a big company. >> I think the question here is how unique will each of our agents look like? And that's what really drives us. When we wake up in the morning, what really drives us is I'm going to create something with my talent, with my expertise, with my hard work that's unique compared to other people. All right, like you're talking about 25 years of your your hard work. And I think the personal agents will allow us to create this highly crafted, you know, something that you're really proud of. And the tools to create your own personal agents are actually going to play a very big role. Uh so, the Ramesh agent, the Nina agent is going to look very very unique. They'll sometimes they'll have economic value, but sometimes it doesn't have economic value, but it's useful in some other ways. And that's the future. That's really the glorious future we can have where we have 8 billion agents that are very unique. Everybody's very proud, very excited about crafting >> a very thought-provoking conversation, and if you have any questions, I think there quite a few people with the Nanda t-shirts on, and they're all really smart, so talk to them. So, thank you very much, Ramesh. >> Thank you, Nina. And

Assistir ao vídeo original no YouTube

Assuntos: Agentes de IA, AGI. Publicado no YouTube em 2026-10-16.

Seleção automática por assunto e idioma. O vídeo pertence ao canal de origem.