Credit: Professor Ramesh Raskar NANDA presentation; watch here
As I sat in the buzzing rooms at MIT last week, surrounded by brilliant minds from academia, startups, enterprise, venture, and beyond, I couldn't help but feel the humbling spark of witnessing something big. I’ve long been amazed at how protocols like TCP/IP and HTTP transformed the internet into a global fabric of connection, not through top-down mandates, but through open collaboration, simplicity, and a relentless focus on utility. Now, with AI agents poised to redefine how we interact with the digital world—abstracting away search workflows, ads revenue models, and even decisions—we're at a similar inflection point. The Web3 Quilt workshops, co-hosted by the Advanced AI Society and MIT Media Lab's Project NANDA, brought us together to brainstorm an interoperability layer that stitches AI agent registries across Web2 and Web3. The goal is to create a more vibrant, open internet more suited for the cambrian moment we are in, fostering agentic rails that empower humanity rather than exploit it. Drawing from this rich exchange, below are key themes that emerged and strike me as mission-critical to ensuring this new era serves us all responsibly and neutrally.
Ask: if you are interested in advancing AI, the Web3 Quilt project or otherwise are opinionated about this cause, I’d love to connect, and we at the Advanced AI Society would love to invite you to join the NANDA project working groups! If you are building in this space, I’d especially like to learn whether the below resonate with you, and what blockers you are encountering to honor these principles whether technically, operationally or legally.
Privacy by Default: Safeguarding Autonomy in an Agentic World
Privacy isn't just a checkbox—it's the foundation of trust in a world where agents will handle our most intimate data, from health records to financial decisions. In our workshops, we repeatedly circled back to confidentiality and privacy-preservation as core human rights*, emphasizing that once lost, it can't be restored. As agents negotiate on our behalf, querying registries and syncing data across networks, any leak could cascade into systemic vulnerabilities, eroding user agency and inviting exploitation. By baking in principles like data minimization, selective disclosure, and unlinkability—compatible with GDPR and similar laws—we create a system where individuals control what’s shared. This is crucial because a vibrant internet of agents must prioritize human autonomy and not conflate human rights with agentic permissions; without privacy-preserving designs, we risk repeating past design errors and agents turning into surveillance tools. Our discussions highlighted tools like zero-knowledge proofs as paths forward, allowing verifiable interactions without exposing raw data, ensuring the quilt becomes a shield for prosocial innovation.
*See Frank McCourt’s and Michael J Casey’s excellent book on this topic: Our Biggest Fight (2024)
Interoperability and Protocol-Agnosticism: Stitching Silos Together
At the heart of the Web3 Quilt is the dream of seamless connectivity, much like how HTTP enabled the web's explosion by being protocol-agnostic. Debates arose about building atop existing standards—A2A, MCP, W3C’s DID, DNS—while filling gaps to connect disparate registries without forcing builders to adopt new frameworks. This theme is vital because fragmentation breeds monopolies; if agents can't "speak" across Web2 and Web3 silos, we'll end up with walled gardens dominated by a few giants, stifling innovation and accessibility. By designing for composability—think modular trust anchors and agnostic key management—we enable agents to discover, handshake, and collaborate effortlessly, from enterprise firewalls to decentralized networks. This is about equity: smaller players and individuals shouldn't be locked out. Our notes underscored that true interoperability, with features like semantic versioning and deterministic IDs, will drive widespread adoption, mirroring TCP/IP's success through simplicity and openness, ultimately fostering a collaborative ecosystem where agents solve global challenges together.
Security and Zero-Trust: Building Verifiable Foundations
In a landscape teeming with trillions of agents, security can't be an afterthought—it's the bedrock that prevents chaos. Trustless models where every claim is proven cryptographically, from tamperproof registries to remote attestation via TEEs (Trusted Execution Environments), are one solution. Hierarchies of trust roots, free from centralized DNS vulnerabilities, can effectively and safely bind identities, keys, and functions, which is crucial because an open internet of agents will attract adversaries; without verifiability, revocation, and censorship-resistance, bad actors could hijack systems, leading to misinformation cascades or economic sabotage. By embracing open-source auditing, quantum-safe cryptography, and modular trust—where users pick their anchors—we mitigate risks like degeneration into "super-agent" monopolies. The conversation hearkened back to email's evolution: early protocols succeeded by balancing openness with security layers. Here, zero-trust empowers safe, high-value use cases, like lifetime personal agents, ensuring the quilt serves humanity's long-term wellbeing over short-term gains in an ironclad way.
Inclusivity and Equitability: Democratizing Access
We can't build the future if it's only for a privileged subset of humanity—more than half of the global population uses email whereas we want to be building rails for all 8 billion+ humans, with an eye to advancing equitable designs that support humanity writ large. Themes like fast resolution, partitionability, and multiple service choices could lower barriers, making the quilt accessible to underrepresented segments of society. This is essential because an agentic internet should serve prosocial needs rather than exacerbate divides. If costs (gas, compute) or complexities exclude smaller developers or regions, we risk recreating or aggravating existing inequities. Optimizing for decentralized economic design, related incentivization mechanisms and human-friendly interfaces are a first step in fostering adoption. Lots of great lessons to be gleaned from crypto’s incentivization experiments here.
Censorship-Resistance and Openness: Preserving Freedom in Distributed Networks
Openness is non-negotiable—transparent, auditable, and distributed systems that resist censorship while allowing upgrades via governance. We envisioned a quilt that's forkable, with curators but no single point of failure, supporting mixed private/public networks and avoiding lowest-common-denominator pitfalls. This theme is critical because agents will mediate commerce, communication, and creativity; censorship could silence innovation or enforce biases, turning the internet into a controlled narrative. By designing for revocability without central choke points—e..g., decentralized consensus on data correctness—we enable resilience against capture. Risks were mentioned like protocol wars extending into agent space, but openness, inspired by TCP/IP's neutral pipes, counters this by promoting novelty and traction. Ultimately, a censorship-resistant quilt nurtures a vibrant ecosystem where agents evolve collaboratively in service of humanity.
Energy Efficiency and Sustainability: Aligning Tech with Planetary Needs
Amid talks of scalability to trillions of agents, energy efficiency also emerged as key—optimizing for least-energy paths, mindful of compute's environmental toll is important. Folks debated tradeoffs in confidential computing costs versus performance, urging designs that prioritize sustainability alongside speed. This is mission-critical because an exploding agent economy could exacerbate climate crises if unchecked; agents should solve global issues, not compound them. By incorporating efficiency into non-functional requirements—like choosing protocols that minimize power use—we ensure the quilt is future-proof, compliant with evolving regs, and prosocial. Reflecting on AI's current footprint, we saw sustainability as a chance to lead responsibly and develop an adoption moat. Sustainable designs will drive ethical adoption, proving that a more open internet of agents can harmonize human progress with planetary health.
As we compile and refine ideas further, I invite you—academics, developers, founders, enterprises, and VCs—to join the conversation. Feedback remains open and working groups are under way, with progress to be announced at Stanford on September 8.
Drop your thoughts below - I’d love to hear from you!