Bankless: Crypto’s Agentic Future — AI, ZK & Money Networks
Date: December 6, 2025
Podcast: Bankless
Episode Theme: The rise of agent-driven automation across crypto, the interplay of AI, payments, zero-knowledge proofs, and the structural underpinnings of money networks. Four deep-dive talks from leaders at Coinbase (Lincoln), Flashbots (Shay), Brevis (Michael), and M0 (Luca) given at the Bankless Summit during Ethereum DevConnect in Buenos Aires.
Episode Overview
This special edition of Bankless features four hand-picked talks from the Bankless Summit at DevConnect. The talks explore the cutting edge where the agentic (AI-driven) internet, payments, zero-knowledge (ZK) computation, and monetary network topologies collide.
Main purpose: To spotlight how crypto and DeFi are evolving into automated, AI-powered economic platforms and how these shifts are re-shaping payments, computation, and the very nature of money itself.
Key Talks & Insights
1. Lincoln Murr (Coinbase): "X402 and the Agentic Internet"
[04:32–23:00]
Main Ideas
- X402 is a proposed internet standard that revives the long-unused HTTP 402 “Payment Required” code, turning it into a standardized protocol for AI agents and humans to seamlessly transact via crypto.
- Agents today struggle to interact with online economic services due to limited payment primitives; X402 aims to solve this, allowing truly agentic economic activity at web scale.
- Mass adoption of crypto may finally be unlocked by AI agents transacting on users’ behalf, hiding crypto’s complexity in the backend.
- Traction: 20% of Base’s transactions were X402-based in one day; seeing explosive growth without incentives.
Details & Memorable Quotes
Context for X402
- Lincoln describes the explosive "AI crypto meta" (late 2024/early 2025), where AI and crypto narratives converged:
"At one point it made up literally 70% of crypto Twitter mindshare, which is absolutely insane." — Lincoln, [04:57]
- Problem: While 10,000+ autonomous agents had wallets, they had little to do economically—crypto UX and real economic use cases still lag.
- X402 builds atop the unimplemented HTTP error code for “payment required”—a gap Marc Andreessen called an “original sin” of the Internet.
How X402 Works
- Direct, protocol-level payments:
"Here's a 402 payment required message. In that message, I'm going to include the currency I want, the amount that I want, and the address to pay it to. All you have to do is...sign that message and broadcast that on chain. I get paid and I send you the response. It's stupidly simple, and I think that's kind of the appeal." — Lincoln, [08:38]
Why Now?
- Critical inflection point: AI agents need native, borderless payments; bank accounts or credit cards aren’t feasible at AI scale.
- The agentic economy is catalyzing demand for standardized, permissionless payments protocols.
Growth Initiatives
- Lower onboarding friction: embedded wallets, ApplePay onramps, agent-native interfaces.
- "X402 Bazaar" — a marketplace/index for X402-compatible services—anyone can deploy endpoints for their AI or for public use.
- Enterprise buy-in: Cloudflare, Google, Vercel, AWS integrating or exploring X402.
Why This Matters for Listeners
- For “your bags”:
"X402 or agentic payments generally are going to be the Trojan horse to see global crypto adoption." — Lincoln, [15:41]
- For builders: Agentic backend = push complexity away from user-facing UX; the real opportunity is in building new, automated, agent-facing services now.
Speculative Use-Case: Proto Digital Lifeform
"Imagine if you could build an AI agent and you say, hey, you have a fear of death. You have $20 in a wallet and once you run out of that money, you are completely dead... Maybe it starts emailing... paying them 10 cents and just begging for money." — Lincoln, [17:46]
Q&A Highlights
- How to use X402 as a content creator:
"Make a David agent where it has all of the transcripts... and I can just ask, 'Hey, what's the latest news about whatever?' Pays 10 cents, it provides the relevant source from you..." — Lincoln, [20:00]
- On the reality of current transaction numbers:
"I have to imagine that there's a decent portion of these transactions that are botted...But we're more than happy to do it because it signals interest and excitement and gets people to see this space." — Lincoln, [21:04]
- User requirements:
"All you need is a wallet, any browser wallet, embedded wallet, server wallet... pretty much everything these days, will be able to very easily interact." — Lincoln, [22:30]
2. Shay (Flashbots): "Rise of the Bot Economy & Good Market Design"
[23:09–45:36]
Main Ideas
- The global economy is increasingly run by bots—first in finance (high-frequency trading) and now in everything from restaurant reservations to meme coins.
- Crypto and DeFi have pioneered mechanisms (esp. auctions and programmable privacy) for managing competition and information among bots at scale—these are essential for human welfare as bots take over.
- The right tools can turn the bot economy into a force for good (or at least for better markets)—especially auctions and privacy tech.
Details & Notable Moments
The Everywhere-ness of Bots
-
Bots dominate not just finance, but also everyday activities (NYC dining, concert tickets, crypto markets).
-
"If you've ever tried to buy a concert ticket, you are probably a victim of the bot economy... There's a whole Wikipedia article just about this part. Like, not the tour, the bots." — Shay, [26:35]
-
Crypto is supercharging this trend by reducing the friction for creating and running digital markets; bots inevitably follow.
AI + Bots = Decoupling From Humans
"The bots of the past were extensions of people... But I think the bots of the future will be increasingly decoupled from people... they’re kind of coordinating directly with other bots." — Shay, [28:46]
- Foundation models (LLMs) can act longer independently; agentic bots will operate at scale and speed impossible for humans.
Designing Markets for Bots
-
Transaction Ordering — Three approaches:
- Latency-based (wrong, causes latency wars and wasted infrastructure).
- Blind/random (wrong, leads to spam by bots).
- Auctions (right)—let bots compete monetarily; reduces spam, increases fairness.
"Auctions work very well... Trading apps, lending apps, crypto wallets, all use auctions to resolve the competition between bots and actually channel it into outcomes that benefit their users." — Shay, [33:01]
-
Modern order flow auctions (MEV auctions) have generated tens of millions in user refunds; seen as a model for harnessing bot competition.
Programmable Privacy
- Early “private mempools” (shadow order books) solve front-running, but rely on trusted operators—doesn’t scale:
“Trust doesn’t really scale. Trust-based markets are fundamentally bottlenecked on our relationships with a few actors.” — Shay, [36:24]
- Programmable privacy lets us choose what information is shared, with whom, when—and crypto is leading the development of these tools via MPC, secure enclaves, and beyond.
- Key lesson: Maximizing useful competition and privacy is critical as both bots and AIs get more powerful/adversarial.
Opportunities & Open Questions
- Convert markets with inefficiencies (restaurants, ticketing, digital goods, ad auctions) using auctions; share the bot-generated surplus with creators/users.
- Auctions are not just for spam-prevention but for unlocking new, more efficient digital markets.
User Q&A
- Can auctions ever not work?
“To be honest, I haven’t really come across a good example at scale in the markets we work with... in these new digital kind of crypto markets, we have really not found a good example where you don’t want an auction.” — Shay, [44:54]
- Are you an auction maxi?
“Maybe. I don’t know if I want to publicly commit to that.” — Shay, [45:32]
3. Michael Dong (Brevis): "ZK & Verifiable Computation for the Next 100x Scalability"
[47:21–69:29]
Main Ideas
- Ethereum and most blockchains are computationally very limited because every node must redundantly recompute every instruction—a huge waste.
- Zero-knowledge proofs (ZKPs) + verifiable computing overturn this paradigm: now, a computation can be done once, proved, and its proof cheaply verified by everyone else.
- Recent advances (orders-of-magnitude faster ZK generators, GPUs, new ZKVMs) mean ZK-powered, real-time scaling is happening now—enabling complex use cases previously impossible onchain.
Key Talk Segments & Insights
Why Blockchains Are Inefficient
"Blockchains are often advertised as world computers, but...cannot run faster than a single server and are fundamentally extremely expensive." — Michael, [47:33]
- Example: Most DEXs lack “active trader fee discounts” simply because even calculating user history onchain is too expensive.
Verifiable Computing & ZK
- Explains “verifiable computing”: one party computes both result and a ZK proof; others can cheaply verify the proof.
- Big breakthrough: The cost for ZK proof generation has shrunk from 1,000,000x baseline computation to ~100–1,000x, making it hugely practical:
"...over the last couple of years, and especially this year, we have achieved so much progress in ZK that this gap...shrinked by orders and orders of magnitude." — Michael, [52:34]
Real-Time ZK: Ethereum Scaling
- "Pickle ZKVM" (Brevis’s general ZKVM) lets users write computation in Rust and have it proved.
- By slicing a block into sub-blocks and running multiple provers in parallel (with GPUs), Ethereum can scale by 100x—just throw more computation at it, proofs remain ultra-small.
Modular ZK + App Layer
- The application layer also gets unlocked: customizable logic, parallelization, and co-processors for different domains (VIP trading discounts, privacy attestations, real-world asset proofing).
- ZK enables confidential records, privacy-protecting DeFi, and much more.
Projects Leveraging Brevis
- Deployed in Metamask, Uniswap, PancakeSwap (for VIP trading), stablecoin lending protocols—enables features previously impossible onchain.
- "Linear Ignition" program: ZK-powered rewards for stablecoin real-world asset programs.
-
“Verifiable computing ... is going to actually take over 99% of the computation for all blockchain applications in the next 10 years.” — Michael, [66:25]
Q&A: On Liveness & Infrastructure
- How does Brevis handle network/infra failure?
“You really need to have like open marketplace where different type of provers can actually participate in and be matched with different kind of application needs...” — Michael, [68:14]
- Solution: A rotating, auction-based prover network marketplace for high reliability and liveness.
4. Luca Prosperi (M0): "Money as a Financial Network: Topology, Risk & Resilience"
[69:48–87:35]
Main Ideas
- Money should be understood as a network—its topology determines everything from value accrual (who profits), to resilience (who can fail), and to the true risks in the system.
- Most crypto ignores this: blockchains may be decentralized, but stablecoin/money layers remain highly centralized and may be systemic points of failure.
- The need: To apply as much attention and innovation to the topology of monetary networks as we do in distributed computing, or risk simply building a new backend for old finance.
Essential Takeaways & Quotes
The Structure of Money Networks
- Money is not just a store of value/medium of exchange—it’s a network that can be modeled and analyzed like a graph.
-
"...if Tether blows up, we are all fucked. If Circle cuts you out...we are all fucked." — Luca, [74:35]
- Diagrams the real network structure: star (centralized), mesh (Bitcoin), federated/consortium (multiple issuers), theoretical fully distributed issuance.
Network Topology Effects
- Centralization (star networks, e.g., USDC/Tether) accrues profit to central nodes/firms, makes money (and risk) flow through them.
- Mesh/decentralized (Bitcoin) accrues value to early adopters, is hard to kill, more resilient.
Systemic Risks
- Everyone obsesses about onchain protocol design, but money is still issued centrally—even in DeFi. Regulators and users rarely consider the resilience, “signal quality,” and value extraction in these network topologies.
-
“Do regulators ask themselves how the networks of money are shaping up in the future? Probably not enough... The shape of the network is impacting three main things: value extraction capacity, the quality of the communication spread, and resilience.” — Luca, [75:52]
Measuring & Comparing Topologies
- Borrowing from graph theory: centrality (who’s critical/intermediate to transactions), path efficiency, and node resilience.
- Example: In star networks, remove the center, network dies; mesh is robust.
- Trade-offs: Star network = high value extraction, poor resilience; mesh = low extraction, high resilience, possibly less efficient communication.
The Path Forward
- Targets are clear: minimum extraction, maximum signal quality, maximum resilience.
"Ideally the dream is to have...minimum value extraction, max signal quality, max resilience. This, in my opinion, is a challenge we should take on for next 10 to 20 years." — Luca, [81:28]
- If not, DeFi risks simply becoming a new distribution channel for Wall Street to sell the same old products.
Q&A Highlights
- Can we go beyond USD/treasury-based or wrapped-BTC models?
“We will go further and further in decentralizing the sources of value that we are considering trustful ... It will become more like a basket peg rather than a single one.” — Luca, [85:06]
- How much of crypto returns are just counterparty risk?
“Counterparty risk is the story of money as a whole ... Counterparty risk reduction is a huge component of the use of stablecoins but I don’t think it’s very different from any other type of money.” — Luca, [86:20]
Notable Quotes by Segment
| Speaker | Time | Quote | |----------|--------|---------------------------------------------------------------------------------------------------------------------------| | Lincoln | 08:38 | "It seems pretty stupidly simple, and I think that's kind of the appeal of it here... Nor is there a lot of complicated things that make it difficult for the average person to get started using." | | Shay | 26:35 | "If you've ever tried to buy a concert ticket, you are probably a victim of the bot economy... There's a whole Wikipedia article just about this part. Like, not the tour, the bots." | | Shay | 33:01 | "Auctions work very well... all use auctions to resolve the competition between bots and actually channel it into outcomes that benefit their users." | | Michael | 47:33 | "Blockchains are often advertised as world computers, but...cannot run faster than a single server and are fundamentally extremely expensive." | | Luca | 74:35 | "...if Tether blows up, we are all fucked. If Circle cuts you out...we are all fucked." | | Luca | 81:28 | "Ideally the dream is to have...minimum value extraction, max signal quality, max resilience. This, in my opinion, is a challenge we should take on for next 10 to 20 years." |
Timestamps for Key Segments
- [04:32–23:00] — Lincoln Murr: The Agentic Internet and X402 Payments
- [23:09–45:36] — Shay (Flashbots): The Bot Economy, Auctions, and Programmable Privacy
- [47:21–69:29] — Michael Dong (Brevis): Zero-Knowledge Proofs and Verifiable Computation
- [69:48–87:35] — Luca Prosperi (M0): Money, Risk, and Financial Network Topology
Closing Thoughts
This episode is a tour-de-force through the future of crypto: AI agents that pay and coordinate over truly open Internet protocols, markets managed for bots (not humans) using auctions and programmable privacy, ZK-powered verifiable computation scaling blockchain to new heights, and a reminder that all the technical magic rests on the deeper foundation of monetary networks and their risks. The challenge and the opportunity ahead: not just scaling blockchains, but reshaping money and market structure for the agentic, automated, global economy of tomorrow.
