
They're Making TCP/IP For AI, And It's Called NANDA
Not too long ago, I realize that there's a new type of standard reinventing the web that hardly anyone is talking about – yet.
Let's start with the name: Networked Agents and Decentralized AI or NANDA is a web protocol for AI agents that's largely being worked on right here at MIT.
Here's an important note – in terms of design context, NANDA builds on Anthropic's Model Context Protocol (MCP) that provides for standardized transactions between AI agents. NANDA adds Internet capability and protocols, so that these agents can 'do things' over the web.
That's it, in a nutshell, but getting there is going to take work. What I found really interesting is that hardly anyone knows that this effort is taking place.
I hadn't heard of NANDA before, so I did a quick web search. I mostly came up with something related to the nursing industry, and alternately, a name for an LLM project geared towards the Hindi language.
However, some of the authors of a paper on NANDA have released items to LinkedIn explaining how this protocol is going to work.
'Just as DNS revolutionized the internet by providing a neutral framework for web access, we need a similar infrastructure for the 'Internet of Agents,'' writes Ayush Chopra, a PhD candidate at MIT who is working on this idea. 'We're launching NANDA - an open protocol for registry, verification, and reputation among AI agents - in collaboration with national labs and global universities (decentralized across 8 time zones) - NANDA will pave the way for seamless collaboration across diverse systems, fully compatible with enterprise protocols like MCP and A2A. This initiative is a step toward democratizing agentic AI, creating an ecosystem where specialized agents can work together to solve complex challenges—just like DNS did for the web.'
As for ChatGPT, when I asked the model 'what is NANDA for the web?' it came back with items about the nursing organization.
I had to specifically ask for the full name of the project, Networked Agents and Decentralized AI, but it did attribute this to the MIT Media Lab, and Ramesh Raskar, a friend and colleague of mine, in particular. It came up with this very quick dual response, which I thought was pretty much on point:
✅ Yes, there is a newer field or research project called NANDA — Networked Agents and Decentralized AI.
✅ It's completely different from the nursing NANDA.
That's a good start (and ChatGPT did provide more) – but we can get more from the people who are actually involved.
Now let's go to Raskar himself speaking at an Imagination in Action event earlier in April.
In it, he explained that NANDA is really a protocol for the Internet, the 'TCP/IP of AI' (I thought that was so good, I made it the title of this piece.'
He talked about how the Internet is for computers, but the World Wide Web is for people. NANDA is for AI agents, he explained, to be able to use the Internet in the same ways as people do.
As an example, Raskar talked about the challenge of planning a birthday party for a child, in which one could conceivably use AI agents to buy cake and balloons and other items, and make all of this process automated.
As for buy-in, he said NANDA has active DNS registration at 15 major universities. He mentioned a three-part process for deployment: 1. host registration, 2. build APIs 3. connect to open source.
Later, Raskar was asked a series of questions by Aaron Pressman, and talked about how this may roll out in the future.
'Everybody's craving something that's open and vibrant,' he said. 'MIT has a history. We are deeply inspired by Tim Berners Lee and the World Wide Web Consortium right here at MIT, you know, that really showed how we can create this open, vibrant web from the early '90s to about '95. And that really unleashed (a lot of) economic value.'
Calling for projects that are 'ambitious, open and inclusive,' Raskar put forth the idea that we are now in a 'wisdom economy,' as opposed to a transaction or knowledge economy.
'We can imagine billions and trillions of dollars of value being created when each one of us has the ability to create our own AI agents, every shop, every individual, every organization, can train and deploy their own agents,' he said, promoting NANDA as a platform to 'build the commerce backbone' that will drive the new Web.
I thought this was interesting: Raskar, in his talk, and Chopra, in his written explanation, introduced us to a new term: A2A – Agent to Agent.
In talking about the use of cryptography and smart contracts, Raskar also gave us 'Know Your Agent' (KYA) presumably as a correlate to the Know Your Customer (KYC) protocols in place in crypto.
'If you think about the internet stack,' he said, 'we left things like Yelp … in the application layer, because (they include) the human messiness of scoring. But because this is agent-to-agent communication, lot of interactions are algorithmic.'
Raskar's presentation helps us to put NANDA in context, and understand what it's for as well as what it's likely to do. I was excited about this partly because it's not really common knowledge yet.
In the beginning, we had Web 1.0 – the read-only web, where people were mostly just looking at text and images coded in HTML.
Web 2.0 was the read/write web, where web users could put information into web forms and send it back to a webmaster or host.
Web 3.0 is the read/write/function web, where people could send information both ways, and/or utilize functions built into web technology.
Web 4.0 is where machines are going to be doing all those things for us, and we're going to sit back and enjoy the music.
At least, that's the idea.
Look for this work to continue and really make its way into the public consciousness rather soon.
Hashtags

Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Yahoo
35 minutes ago
- Yahoo
Citizens JMP Raises DoorDash Price Target, Highlights AI Advertising Growth
DoorDash Inc. (NASDAQ:DASH) ranks among the best consumer discretionary stocks to buy now. On June 12, Citizens JMP maintained its Market Outperform rating on DoorDash Inc. (NASDAQ:DASH), while increasing the price target to $235 from $225. Photo by Jon Tyson on Unsplash The increase comes follows DoorDash, Inc. (NYSE:DASH) acquiring Symbiosys, an AI-powered advertising platform, for $175 million and introduced new AI-powered advertising solutions for merchants and brands. Additionally, DoorDash disclosed that its advertising division generated $1 billion in revenue last year, indicating the company's strategic emphasis on growing this revenue stream. According to Citizens JMP, DoorDash's global advertising income in 2024 was over $950 million, reflecting a 1.2% penetration rate, though this remains less than the 2% level attained by rival Uber's Eats restaurant division. One of the largest online food delivery companies in the United States, DoorDash, Inc. (NYSE:DASH) operates a delivery platform based in San Francisco, California. Historically, the company has categorized itself into five business segments: platform services, advertising, non-restaurant services, international restaurants, and US restaurants. While we acknowledge the potential of DASH as an investment, we believe certain AI stocks offer greater upside potential and carry less downside risk. If you're looking for an extremely undervalued AI stock that also stands to benefit significantly from Trump-era tariffs and the onshoring trend, see our free report on the best short-term AI stock. Read More: and Disclosure: None. Error in retrieving data Sign in to access your portfolio Error in retrieving data Error in retrieving data Error in retrieving data Error in retrieving data
Yahoo
35 minutes ago
- Yahoo
This AI-powered startup studio plans to launch 100,000 companies a year — really
Henrik Werdelin has spent the last 15 years helping entrepreneurs build big brands like Barkbox through his startup studio Prehype. Now, with his new, New York-based venture Audos, he's betting that AI can help him scale that process from 'tens' of startups a year to 'hundreds of thousands' of aspiring business owners. The timing certainly feels right. Mass layoffs across a variety of industries have left many workers reconsidering their career paths, while AI tools have markedly lowered the barrier to building digital products and services. At the center of that Venn diagram is Werdelin's latest venture, with its promise to help 'everyday entrepreneurs create million dollar AI companies' without requiring technical skills. Werdelin's journey from Prehype to Audos reflects the broader transformation happening in entrepreneurship right now. At Prehype, the focus was on working with tech founders to build traditional startups, the kind that might raise millions and aim for billion-dollar exits. Now, he tells TechCrunch, 'What we're trying to do is take all that knowledge, all the methodology that we've created over the years of building all these big companies, and really trying to democratize it.' The idea is that 'everyday entrepreneurs' may sense a shift is afoot but may not be keen to experiment with so-called AI agents or know how to reach customers. Audos is more than happy to help them, supplying these individuals with AI tools to build sophisticated products using natural language, and taking advantage of social media algorithms to find them their niche customers. 'Facebook and a lot of these platforms, they are just incredible algorithms, and they're incredible at figuring out [how to reach your customer] if you define a customer group,' says Werdelin, who co-founded Audos with his Prehype partner Nicholas Thorne. In fact, Audos uses this system to quickly test whether a founder's business idea has sustainable customer acquisition costs. The approach seems to be working. Audos has helped launch 'low hundreds' of businesses since its beta launch, with its own customers discovering the platform through Instagram ads asking 'Have you ever thought about starting something, but don't know where to go?' Among them, Werdelin says, are a car mechanic who wants to help people evaluate repair quotes, an individual who is selling 'after death logistics' services, virtual golf swing coaches, and AI nutritionists. In a winking reference to billion-dollar businesses, or so-called unicorns, he calls these one- and two-person teams 'donkeycorns.' All went through the same process: they clicked on Audos's ad, its AI agent launched a conversation to figure out the problems these individuals want to tackle and who they want to serve, and, when it was satisfied with the answers, Audos got them in front of potential customers as fast as possible. As for returns, Audos operates on a fundamentally different model than traditional accelerators or venture capital. Instead of taking equity, the company takes a 15% revenue share from the businesses it helps launch. In return, founders get up to $25,000 in funding, access to those AI-powered business development tools, and help with distribution, primarily through paid social media advertising. 'We're not taking any equity in their business,' Werdelin says, partly because 'we don't think these companies might ever get sold. What we're really inspired by are the mom-and-pop shops that are the backbone of our society.' The revenue share continues indefinitely, similar to platform fees charged by Apple's App Store. For founders, that means giving up a significant portion of their revenue in perpetuity — a 15% cut that could cost entrepreneurs hundreds of thousands of dollars over time. Some will undoubtedly see that trade-off as worthwhile; others might question whether the long-term costs justify the benefits. Audos's value proposition raises other questions given how quickly the landscape is changing. While Werdelin emphasizes helping founders build relationships with customers, it's unclear how much of that work the AI agents can actually handle. There's also the matter of differentiation. As Werdelin readily acknowledges, 'the world is full of these tools' and they're getting better rapidly. What happens when entrepreneurs can access similar AI capabilities without paying a permanent revenue tax? Audos's VCs don't sound worried about those scenarios. True Ventures led Audos's $11.5 million seed round, with partner Tony Conrad explaining the appeal in a Zoom call this week. In addition to having confidence in Werdelin and Thorne, says Conrad, 'I think there are just lots and lots of people' who might eagerly embrace the opportunity to work with a platform like Audos. Conrad draws parallels to Instagram's $1 billion exit with just 13 employees, suggesting that AI could enable even more leverage, even if Audos — which itself employs just five people altogether currently — isn't chasing unicorns. As Werdelin explains it, 'What we're after here is the millions of people who can create million-dollar businesses or half-million dollar businesses that are real and life changing.' Adds Werdelin separately of why he spun up Audos, 'What we're trying to do is to figure out how you make a million companies that do a million dollars turnover. That's a trillion dollar turnover business.' It doesn't sound crazy. Extending the benefits of entrepreneurship to people who traditionally haven't had access to startup capital or technical skills is an increasingly compelling proposition as traditional employment begins to feel less and less stable. 'We believe that there should be somebody who goes out and really helps these smaller entrepreneurs that are building something that is not venture backable,' says Werdelin. 'We believe that the world is better with more entrepreneurship.' Audos's other investors include Offline Venture and Bungalow Capital, along with numerous high-profile angel investors – Niklas Zennstrom and Mario Schlosser among them. Pictured above, left to right, Audos co-founders Nicholas Thorne and Henrik Werdelin.
Yahoo
an hour ago
- Yahoo
Windows Recall Is Getting Its First Major Update
Windows Recall, a "privacy-focused" AI tool for Copilot+ PCs designed to make it easier to retrace your digital steps if you forgot what you did the day before, is going to receive a major update: a homepage. Recall has been mired in controversy since it was announced, as early critics were quick to raise concerns that the feature could be an enormous security and privacy vulnerability. Since then, Microsoft has gradually been rolling out changes and improvements to Recall designed to alleviate some of those concerns and make Recall more user-friendly. Microsoft's latest addition—which is currently in both the Beta and Dev channels of the Microsoft Insider Program—is a homepage, which Microsoft has dubbed a "personalized productivity portal." As user-facing additions go, a homepage is a pretty big improvement. As it stands currently, Recall's primary interface is a basic timeline that you can scrub through, or you can use the search bar. The search bar is a "semantic search," which means it tries to intelligently match the intention of your search, not just the actual keywords you type. The new homepage prominently features the three apps you used the most in a 24-hour period, plus the websites you've spent the most time on. It'll also display your snapshots, which should make it easier to find what you're looking for, rather than manually scrubbing through the timeline. The homepage also features a new navigation bar on the left that integrates some buttons you'd probably expect: a homepage button, a timeline button (which just takes you to the current Recall timeline as it exists now), a Settings button, and a feedback button. They're not particularly exciting features, but they are important for usability. Having the Settings front and center is a particularly welcome addition, since it will be a bit of a reminder that you can prevent Recall from capturing snapshots of certain apps or sites if you don't want them showing up in your history. While the changes being tested are certainly an improvement, I'm not completely sold on the utility unless it brings something really special to the table. If you're working on your PC day in and day out, you already know which applications you use a lot, and PowerToys (a collection of software distributed by Microsoft) already includes WorkSpaces, which lets you open a preset list of programs, specific locations on your PC, or websites with the press of a few buttons. Browsers also already list the sites you've visited recently and the sites you visit most frequently. In many ways, Recall's present form feels a lot like a solution—and a reason to use cool new AI tricks—in search of a problem.