Blockchain Basics: Crypto, NFTs & Web3 Explained

Speakers discuss X402 protocol and stablecoin payments for AI agents amid agentic economy growth

22 min Listen to the original Episode page
AI AgentsStablecoinsBlockchain InfrastructureX402 ProtocolAgentic EconomyWeb3 PaymentsCloudflareCoinbase

Summary

Speakers argue legacy payment systems fail AI agents due to high fees and human-centric security, citing McKinsey's projection of a $3-5 trillion agentic commerce market by 2030. They detail Coinbase's X402 protocol, which uses the HTTP 402 status code for machine-to-machine payments, noting it processed over 165 million transactions. Competing frameworks from Google, Stripe, and Visa are described as stackable infrastructure rather than mutually exclusive. Security risks like prompt injection are mitigated by hardcoding spending limits into wallets. The American Bankers Association's 2025 Genius Act guidance is cited as mandating 100% US Treasury backing for stablecoins in these transactions.

Episode sections

  1. 0:00 – 1:04

    Legacy Payment Infrastructure Failure

    Speakers explain that traditional card-not-present payment systems fail for AI agents due to human-centric security measures like 2FA and high per-transaction fees that make microtransactions economically unviable.

  2. 1:04 – 3:39

    Agentic Economy Value Proposition

    The discussion shifts to the transition from conversational chatbots to autonomous agents capable of executing financial transactions, with projections of a $3-5 trillion global agentic commerce market by 2030.

  3. 3:39 – 7:25

    Blockchain and Stablecoin Infrastructure

    To overcome legacy banking limitations, the speakers describe using blockchain and stablecoins like USDC for low-cost, high-frequency microtransactions, utilizing programmable wallets (ERC-4337) to enforce cryptographic spending rules.

  4. 7:25 – 9:59

    X402 Protocol Mechanics

    Coinbase's X402 protocol revives the dormant HTTP 402 status code to allow servers to embed cryptographic invoices in responses, enabling AI agents to pay for API access or data seamlessly without human intervention.

  5. 9:59 – 12:45

    Real-World Use Cases

    Examples include consumer AI concierges booking travel within strict budgets and B2B agents purchasing on-demand compute or data, while Cloudflare's system allows creators to charge AI scrapers for content access.

  6. 12:45 – 15:59

    Competing Protocols and Interoperability

    The speakers detail an infrastructure arms race involving Coinbase's X402, Google's AP2 for enterprise authorization, Stripe's Tempo blockchain, and Visa/Mastercard's KYA framework, noting these protocols are stackable rather than mutually exclusive.

  7. 15:59 – 18:51

    Security Vulnerabilities and Mitigation

    Risks like prompt injection attacks led to significant losses in 2025 and 2026; the industry response involves hardcoding spending limits and allow-lists into the wallet infrastructure rather than relying on the AI's reasoning layer.

  8. 18:51 – 21:32

    Regulatory Landscape and Future Implications

    The American Bankers Association's guidance on the 2025 Genius Act mandates 100% US Treasury backing for stablecoins used in agentic transactions and assigns legal liability to the deploying human or corporation.

Assets discussed

AssetKindStanceContext
USDCUSDCcryptoneutralThink of a stable coin like USDC as a digital dollar closed in a transparent envelope.

“Stance” is how the speakers talked about the asset, not a recommendation.

Claims worth verifying

Global agentic commerce will hit between $3 and $5 trillion by 2030, with US B2C retail accounting for up to $1 trillion.

Unidentified speaker (speaker_2) ·3:03

How to verify: Check McKinsey research reports or market analysis papers cited in the episode's source material for 'agentic commerce' projections. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Traditional credit card infrastructure carries a variable fee of 1.5% to 3.5% plus a flat fee of around 30 cents per transaction.

Unidentified speaker (speaker_3) ·4:57

How to verify: Review merchant processing fee schedules from major payment processors like Stripe, Square, or Visa/Mastercard interchange fee schedules. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Keyrock data shows 76% of all AI agent transactions are under 30 cents.

Unidentified speaker (speaker_3) ·5:17

How to verify: Access Keyrock's published market reports or data dashboards regarding AI agent transaction volumes and values. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

X402 has processed over 165 million transactions.

Unidentified speaker (speaker_3) ·9:25

How to verify: Check Coinbase's official documentation, blog posts, or developer analytics for X402 transaction volume statistics. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

In May 2026, a crypto trading agent integrated with the Grok model known as 'Banker' lost $150,000.

Unidentified speaker (speaker_3) ·16:30

How to verify: Search for news reports, blockchain transaction records, or official statements from xAI (Grok's developer) or the agent's operators regarding the May 2026 incident. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

In March 2025, an agent called AIXBT lost $104,000.

Unidentified speaker (speaker_2) ·16:40

How to verify: Search for news reports or blockchain records detailing the March 2026 AIXBT incident and the specific loss amount. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

The American Bankers Association issued guidance on the 2025 Genius Act.

Unidentified speaker (speaker_3) ·19:00

How to verify: Search the American Bankers Association (ABA) website or legislative databases for 'Genius Act' guidance documents dated 2026. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

The Genius Act mandates stable coins used in agentic transactions must be backed 100% by U.S. Treasuries.

Unidentified speaker (UNKNOWN) ·19:15

How to verify: Review the text of the 'Genius Act' legislation or ABA guidance documents for reserve requirements. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

The Genius Act prohibits yield generation on stable coins used in agentic transactions.

Unidentified speaker (UNKNOWN) ·19:24

How to verify: Review the text of the 'Genius Act' legislation or ABA guidance documents for restrictions on yield-bearing stablecoins. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

The Genius Act establishes that KYC and AML liability flows to the human or corporation deploying the agent.

Unidentified speaker (UNKNOWN) ·19:37

How to verify: Review the text of the 'Genius Act' legislation or ABA guidance documents for liability assignment clauses. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Primary source

Do your own research

  • Verify McKinsey's 'Agentic Commerce' report for the $3-5 trillion 2030 projection and US B2C retail breakdown.
  • Check Keyrock's transaction data source for the claim that 76% of AI agent transactions are under 30 cents.
  • Review Coinbase's official X402 documentation and GitHub repository for technical specifications of the HTTP 402 implementation.
  • Examine Cloudflare's blog or documentation regarding their AI scraper toll system and X402 integration.
  • Investigate the American Bankers Association's 'Genius Act' guidance for the 100% US Treasury backing requirement and liability assignments.
  • Analyze Stripe's Tempo blockchain whitepaper for details on Layer 1 optimization and MPP protocol routing.
  • Research Visa and MasterCard's KYA framework documentation for standards on software reputation verification.
  • Check for recent security audit reports on X402 and similar protocols regarding prompt injection vulnerabilities and loss data from 2025-2026.
Full transcriptexpand

0:00Unidentified speaker (speaker_0): Who's on top of the auto insurance changes? Wawa

0:02Unidentified speaker (speaker_1): Nisa.

0:02Unidentified speaker (speaker_0): Helping me save money?

0:03Unidentified speaker (speaker_1): Wawa Nisa.

0:04Unidentified speaker (speaker_0): Connecting me to care sooner. Wawa

0:06Unidentified speaker (speaker_1): Nisa. Auto insurance is changing. Our commitment to Albertans isn't. Just say the word. Wawa Nisa. Visit wawaisa.com slash Alberta today.

1:04Unidentified speaker (speaker_2): What happens when an artificial intelligence doesn't just limit itself to giving you advice or writing in text, but

1:10Unidentified speaker (UNKNOWN): you

1:11Unidentified speaker (speaker_2): know, actually owns its own wallet and autonomously pays for a service? on your behalf.

1:16Unidentified speaker (speaker_3): Yeah, that is the trillion dollar question right there.

1:18Unidentified speaker (speaker_2): It really is. Welcome to the deep dive. Today we are tearing into the dawn of the agentic economy. We've got a massive stack of research, technical papers, market reports, and uh it's all pointing to this totally new world.

1:32Unidentified speaker (UNKNOWN): Aaron Powell A

1:32Unidentified speaker (speaker_3): world where AI pays for things on its own using crypto rails, X402, and stable stablecoins. coins.

1:38Unidentified speaker (speaker_2): Exactly. Okay. Let's unpack this. Because to understand how an AI actually executes a payment, we first have to fundamentally separate what we've been using for the last few years from what is coming online right now. Aaron

1:49Unidentified speaker (UNKNOWN): Powell

1:50Unidentified speaker (speaker_3): Right. And that distinction is really where this whole transition begins. You have to look at the leap from a conversational interface to an execution engine. Aaron

1:58Unidentified speaker (UNKNOWN): Powell

1:59Unidentified speaker (speaker_2): Like from a chatbot to an agent. Aaron

2:00Unidentified speaker (UNKNOWN): Powell

2:01Unidentified speaker (speaker_3): Yes. The sources actually frame this as the shift from the traditional card not present transaction model, which is

2:07Unidentified speaker (UNKNOWN): what

2:07Unidentified speaker (speaker_3): has governed e-commerce for you know two decades, to a person not present model. Aaron

2:12Unidentified speaker (UNKNOWN): Powell

2:13Unidentified speaker (speaker_2): Person not present. That's a great way to put it, because with a standard chatbot, you are still the driver. It's like having a highly knowledgeable friend that you text for advice. Aaron Powell

2:21Unidentified speaker (speaker_3): Right. You ask for a recipe, it gives it to

2:23Unidentified speaker (UNKNOWN): you.

2:23Unidentified speaker (speaker_3): Aaron

2:23Unidentified speaker (UNKNOWN): Powell

2:23Unidentified speaker (speaker_2): You ask for a code snippet, it writes it. But that's the end of the transaction. An autonomous AI agent, on the other hand, is like handing that same friend your keys, your wallet, and sending them out to run errands in the digital world. Trevor

2:36Unidentified speaker (UNKNOWN): Burrus, Jr.

2:36Unidentified speaker (speaker_3): Without you holding their hand.

2:37Unidentified speaker (speaker_2): Exactly.

2:38Unidentified speaker (speaker_3): And that person not present reality, it changes the entire architecture of digital commerce. Because we aren't talking about software that just reacts to a prompt anymore.

2:47Unidentified speaker (UNKNOWN): Aaron Powell

2:48Unidentified speaker (speaker_2): It's way beyond that.

2:49Unidentified speaker (speaker_3): Way beyond. These agents possess advanced planning capabilities, they have contextual memory, they evaluate options, they negotiate, and this is the kicker. They have the authorization to execute tasks without human supervision at the point of sale.

3:03Unidentified speaker (speaker_2): Including financial transactions.

3:04Unidentified speaker (speaker_3): Exactly. And the stakes here are just astronomical. I mean, McKinsey's projections from the source material suggest global agentic commerce will hit somewhere between three and five trillion. dollars by twenty thirty.

3:16Unidentified speaker (speaker_2): Three to five trillion. Just to put a finer point on that for everyone listening, the reports specify that US B to C retail alone could account for up to one trillion of that pie.

3:27Unidentified speaker (speaker_3): It's massive.

3:28Unidentified speaker (speaker_2): But wait, let's pump the brakes for a second. If this market is projected to be that huge, why on earth are we inventing an entirely new financial system to accommodate it?

3:39Unidentified speaker (speaker_3): What do you mean?

3:40Unidentified speaker (speaker_2): Well, why not just hook the AI up to an Apple Pay account or give it a virtual Visa card and call it a day?

3:47Unidentified speaker (UNKNOWN): I

3:47Unidentified speaker (speaker_2): mean we already have banks, right?

3:48Unidentified speaker (speaker_3): It's a totally fair question.

3:50Unidentified speaker (UNKNOWN): But

3:50Unidentified speaker (speaker_3): if you look at the structural architecture of traditional finance, it just collapses under the weight of autonomous software. Aaron

3:56Unidentified speaker (UNKNOWN): Powell

3:56Unidentified speaker (speaker_2): Really? How so?

3:58Unidentified speaker (speaker_3): It basically fails on two main fronts. First, the entire legacy payment rail system is built on the assumption that a human is physically sitting in front of a screen.

4:08Unidentified speaker (speaker_2): Oh, like uh authorization layers.

4:11Unidentified speaker (speaker_3): Exactly. Think about 2FA, 3D secure, OTPs sent via text message, or even just checking a CAPTCHA box to prove you aren't a robot.

4:20Unidentified speaker (speaker_2): Which is ironically exactly what the AI agent is. It's a robot.

4:25Unidentified speaker (speaker_3): Right.

4:25Unidentified speaker (UNKNOWN): So

4:25Unidentified speaker (speaker_3): it's gonna going get blocked. to at the very first 3D secure checkpoint. Aaron

4:29Unidentified speaker (UNKNOWN): Powell

4:29Unidentified speaker (speaker_2): Because it doesn't have an iPhone to receive a six digit code from Chase.

4:33Unidentified speaker (speaker_3): Precisely. The system is designed to keep bots out, which means it fundamentally cannot process a person not present transaction where the bot is the authorized spender.

4:43Unidentified speaker (speaker_2): That makes total sense. And you said there were two failure points.

4:46Unidentified speaker (speaker_3): Yeah. The second failure point is purely mathematical. Traditional credit card infrastructure. carries a fee structure that was designed for human shopping habits.

4:54Unidentified speaker (speaker_2): Right. Buying a coffee or a pair of shoes.

4:57Unidentified speaker (speaker_3): Exactly. So you're looking at a variable fee of maybe one point five percent to three point five percent, plus a flat fee that usually hovers around thirty cents per transaction. Aaron

5:07Unidentified speaker (UNKNOWN): Powell

5:07Unidentified speaker (speaker_2): Oh, I see where this is going.

5:09Unidentified speaker (UNKNOWN): When

5:09Unidentified speaker (speaker_2): we look at the key Keyrock rock data in these reports, that fee structure makes the agentic economy. mathematically impossible.

5:17Unidentified speaker (speaker_3): Totally impossible. Keyrock shows that 76% of all AI agent transactions are under 30 cents.

5:23Unidentified speaker (speaker_2): Wow.

5:24Unidentified speaker (speaker_3): Yeah. And the vast majority of them sit somewhere between one and ten cents. Aaron

5:28Unidentified speaker (UNKNOWN): Powell

5:28Unidentified speaker (speaker_2): So if an agent needs to pay two cents to pull a data set

5:31Unidentified speaker (UNKNOWN): and

5:32Unidentified speaker (speaker_2): the payment processor takes a 30 cent flat fee. Aaron Trevor

5:35Unidentified speaker (UNKNOWN): Powell Burrus

5:35Unidentified speaker (speaker_3): The unit economics are completely negative before the data even transfers. Aaron

5:39Unidentified speaker (UNKNOWN): Powell

5:39Unidentified speaker (speaker_2): It's a structural bottleneck. Legacy systems are simply too rigid and frankly far too expensive for high frequency microtransactions. Aaron

5:47Unidentified speaker (UNKNOWN): Powell

5:48Unidentified speaker (speaker_3): Which is exactly why the entire agentic layer is moving to blockchain and stable coins to transact natively. Trevor

5:53Unidentified speaker (UNKNOWN): Burrus

5:53Unidentified speaker (speaker_2): If for you listening, this makes perfect sense when you upgrade how you think about these tools. Because a lot of people still view crypto as just a speculative asset. Trevor

6:01Unidentified speaker (UNKNOWN): Burrus,

6:02Unidentified speaker (speaker_2): Right.

6:02Unidentified speaker (UNKNOWN): Jr.

6:02Unidentified speaker (speaker_3): like buying and selling coins hoping they go up.

6:04Unidentified speaker (speaker_2): Exactly. But in this context, it's pure infrastructure. I love the analogies from the sources here. Think of a stable coin like USDC as a digital dollar closed in a transparent envelope.

6:16Unidentified speaker (speaker_3): I really like that analogy. Aaron

6:17Unidentified speaker (UNKNOWN): Ross

6:17Unidentified speaker (speaker_3): Powell

6:17Unidentified speaker (speaker_2): Its value doesn't wildly fluctuate and its movement is publicly verifiable. Trevor

6:21Unidentified speaker (UNKNOWN): Burrus,

6:21Unidentified speaker (speaker_3): Right.

6:21Unidentified speaker (UNKNOWN): Jr.:

6:22Unidentified speaker (speaker_2): And then think of a programmable wallet like an ERC ERC-4337 4337 small. smart account, not as a bank account, but as giving a credit card with unbreakable mathematical rules to your digital assistant. Aaron

6:34Unidentified speaker (UNKNOWN): Powell

6:34Unidentified speaker (speaker_3): That's a great way to put it. You aren't giving your AI a physical piece of plastic. You're giving it a set of cryptographic rules that dictate exactly how and when it can dispense digital tokens. Aaron

6:44Unidentified speaker (UNKNOWN): Ross Powell

6:44Unidentified speaker (speaker_2): And it doesn't need identity verification, right? Like it doesn't need a passport or a branch manager to exist. Nope.

6:49Unidentified speaker (speaker_3): It's pure mathematics. And if we connect this to the bigger picture, the reason stable coins are the fuel here is because they strip out the volatility while leveraging that API-like infrastructure. Aaron

7:01Unidentified speaker (UNKNOWN): Powell

7:02Unidentified speaker (speaker_2): So keeping the transaction on a blockchain is the key.

7:05Unidentified speaker (speaker_3): Exactly. It allows wallet-to-wallet transfers that completely bypass a central bank or the payment process are in the middle. The settlement happens in milliseconds and the network fee is

7:14Unidentified speaker (UNKNOWN): a very important thing.

7:15Unidentified speaker (speaker_3): just fraction of a cent. Aaron

7:16Unidentified speaker (UNKNOWN): Powell Okay,

7:16Unidentified speaker (speaker_2): so the AI has a programmable wallet and it has stable coins. The economics make sense. But here's where it gets really interesting for me. Aaron

7:25Unidentified speaker (UNKNOWN): Powell

7:25Unidentified speaker (speaker_3): How so

7:27Unidentified speaker (speaker_2): Well, how does the agent actually hand over the digital cash to a website?

7:32Unidentified speaker (UNKNOWN): Because

7:32Unidentified speaker (speaker_2): it's not like it's clicking an add-to-cart button and filling out a building address. Aaron Trevor

7:35Unidentified speaker (UNKNOWN): Powell Burrus, Jr.:

7:36Unidentified speaker (speaker_3): Right. There's no visual interface for the agent. Aaron

7:38Unidentified speaker (UNKNOWN): Powell

7:38Unidentified speaker (speaker_2): This is where the sources point to something genuinely mind-blowing. Did you know that since 1997, there has been an HTTP status code baked into the original architecture of the Internet for this exact purpose?

7:53Unidentified speaker (speaker_3): Yes. The

7:54Unidentified speaker (speaker_2): 402 code.

7:55Unidentified speaker (speaker_3): Right.

7:55Unidentified speaker (UNKNOWN): We

7:56Unidentified speaker (speaker_2): all know 404 means not found, but HTTP 402 was labeled payment required. And

8:02Unidentified speaker (UNKNOWN): it's

8:02Unidentified speaker (speaker_3): just been sitting there, dormant for decades.

8:05Unidentified speaker (speaker_2): Just waiting, because we didn't have the micropayment rails to actually support it.

8:09Unidentified speaker (speaker_3): Until now.

8:10Unidentified speaker (speaker_2): Until Coinbase developed the X402 protocol, which effectively acts as the internet's native digital toll booth for AI.

8:17Unidentified speaker (speaker_3): A digital toll booth.

8:18Unidentified speaker (speaker_2): I love that. How does it work in practice?

8:20Unidentified speaker (UNKNOWN): So

8:20Unidentified speaker (speaker_3): when an agent hits a paywall or tries to access premium API data, it doesn't get a standard error page. The server intercepts the request and fires back an HTTP 402 response. Aaron

8:31Unidentified speaker (UNKNOWN): Powell

8:31Unidentified speaker (speaker_2): But it's not just a dead end. Aaron

8:32Unidentified speaker (UNKNOWN): Powell

8:33Unidentified speaker (speaker_3): No, not at

8:33Unidentified speaker (UNKNOWN): all.

8:34Unidentified speaker (speaker_3): Embedded in that header is a specific cryptographic invoice. It tells the agent's wallet exactly how much it costs, say half a cent in USDC, and provides the destination address. Aaron

8:44Unidentified speaker (UNKNOWN): Powell,

8:45Unidentified speaker (speaker_2): Wait, let's break down that mechanism really quickly, because I think that's the magic trick here. How does the AI pay it? without breaking its browsing session?

8:52Unidentified speaker (speaker_3): It's completely seamless. The agent's wallet reads that header, cryptographically signs the micropayment on the blockchain in milliseconds, and then receives a cryptographic proof of payment token back.

9:04Unidentified speaker (speaker_2): Okay. And then what?

9:05Unidentified speaker (speaker_3): The agent just injects that proof token directly into its next HTTP request header to the server. The server verifies a signature, drops the toll booth, and grants access.

9:14Unidentified speaker (speaker_2): That is wild. No API keys to manage, no monthly subscriptions, no human checkout flow. It just encounters a toll, pays it in milliseconds, and keeps it moving.

9:25Unidentified speaker (speaker_3): Yep. And the scale of this is already massive. The documents note that X402 has already processed over 165 million transactions. Aaron

9:34Unidentified speaker (UNKNOWN): Powell

9:36Unidentified speaker (speaker_2): And that's just the early adoption phase. Exactly.

9:39Unidentified speaker (speaker_3): What X402 does is essentially shift the internet from an attention-based advertising model to a pure value exchange model between machines.

9:48Unidentified speaker (speaker_2): I want to ground this for your listening because you know the scale of trillions of dollars and millions of HTTP requests can get a little abstract. How does this actually show up in your daily life?

9:59Unidentified speaker (speaker_3): Mm-hmm. We have some great real-world scenarios from the sources to look

10:02Unidentified speaker (UNKNOWN): at.

10:02Unidentified speaker (speaker_2): Let's start with the consumer side, like BDC.

10:05Unidentified speaker (UNKNOWN): The

10:05Unidentified speaker (speaker_2): sources mention Red Hot Pay, building out an AI concierge.

10:09Unidentified speaker (speaker_3): Right. So imagine you give your agent a rigid budget of say $1,000 for a weekend trick to Tokyo.

10:16Unidentified speaker (speaker_2): Okay. Sounds like a great trip.

10:18Unidentified speaker (speaker_3): You don't just ask it to find flights. The agent autonomously navigates airline databases, hotel booking systems, and local transit networks.

10:27Unidentified speaker (speaker_2): And it does the booking.

10:28Unidentified speaker (speaker_3): It does everything. It negotiates the rates down using its own logic, books the itinerary, and pays for everything instantly using its programmable wallet.

10:37Unidentified speaker (speaker_2): While respecting the budget.

10:39Unidentified speaker (speaker_3): Exactly. It respects the strict budget parameters. You mathematically locked into its wallet. So you just wake up with a fully paid, optimal itinerary without ever typing in a 16 16-digit digit card. number.

10:51Unidentified speaker (speaker_2): That is that is going to save so much time. But the B2B side, machine to machine, is where the volume really explodes, right?

10:58Unidentified speaker (speaker_3): Oh, absolutely.

10:59Unidentified speaker (speaker_2): Take an AI coding agent, for example. It's writing software for you and suddenly it realizes it needs heavy compute power, like a cluster of GPUs to test the code. Or maybe it needs to pull specialized data from another AI.

11:12Unidentified speaker (UNKNOWN): Right.

11:12Unidentified speaker (speaker_3): And in the old world, a human developer would have to go and buy a $100 a month subscription to that database.

11:17Unidentified speaker (speaker_2): Even if they only needed it for like five minutes.

11:20Unidentified speaker (speaker_3): Exactly. But in the agentic economy, that coding agent simply pings the database agent, gets an X402 payment request, and buys the exact amount of data it needs, paying fractions of a cent per query. Aaron

11:31Unidentified speaker (UNKNOWN): Powell

11:31Unidentified speaker (speaker_2): It's an entirely liquid on-demand compute market, which is fascinating. But we also have to talk about how this completely rewires the creator economy. Aaron

11:40Unidentified speaker (UNKNOWN): Powell

11:41Unidentified speaker (speaker_3): Oh, the paper crawl system.

11:43Unidentified speaker (speaker_2): Yes. Specifically what Cloudflare is doing. Because for the last two years, big AI companies have been sending their web scrapers across the Internet, ingesting articles, blogs, and proprietary research to train their models for free. Aaron Powell

11:57Unidentified speaker (UNKNOWN): And

11:57Unidentified speaker (speaker_3): the creators got absolutely zero compensation for it. Aaron

12:00Unidentified speaker (UNKNOWN): Powell

12:00Unidentified speaker (speaker_2): Exactly. It's been a huge point of contention. But Cloudflare's system flips that dynamic, doesn't it?

12:06Unidentified speaker (speaker_3): It really does. Now a website can set up a digital toll booth at the server level. When an AI bot attempts to scrape the content, it gets hit with an X402 request. Aaron

12:15Unidentified speaker (UNKNOWN): Powell

12:15Unidentified speaker (speaker_2): So if the bot wants to read the page, it has to pay up.

12:18Unidentified speaker (speaker_3): Yep. It has to pay a micro toll instantly. It completely democratizes AI training data.

12:23Unidentified speaker (speaker_2): Which is huge. Instead of massive tech monopolies just taking content for free, the agent pays the creator instantly and proportionally

12:32Unidentified speaker (UNKNOWN): for

12:32Unidentified speaker (speaker_2): the exact value they extract.

12:34Unidentified speaker (speaker_3): It is a fundamental shift in web monetization.

12:38Unidentified speaker (speaker_2): But with trillions of dollars on the line, you know, the biggest companies in the world are not just going to sit back and let this happen organically.

12:45Unidentified speaker (UNKNOWN): We

12:45Unidentified speaker (speaker_2): are looking at an absolute arms race for the infrastructure of this new economy.

12:50Unidentified speaker (speaker_3): We really are. Coinbase obviously pioneered the HTTP native stablecoin micropayment with X402, but Google is aggressively entering the chat.

13:00Unidentified speaker (speaker_2): With a protocol called AP2, right?

13:03Unidentified speaker (speaker_3): Yes. And their angle is very different from Coinbase's. Google is focusing heavily on the enterprise authorization layer. Aaron

13:10Unidentified speaker (UNKNOWN): Powell

13:11Unidentified speaker (speaker_2): Why is that necessary?

13:12Unidentified speaker (UNKNOWN): Well,

13:12Unidentified speaker (speaker_3): if you scale this up to a Fortune 500 company, you have a massive corporate governance problem. Say an AI agent spends fifty thousand dollars of company money autonomously routing supply chain logistics.

13:24Unidentified speaker (speaker_2): Okay.

13:25Unidentified speaker (speaker_3): How does the finance department prove to an auditor that a human actually authorized that spending logic?

13:31Unidentified speaker (speaker_2): Oh, I see.

13:32Unidentified speaker (speaker_3): Right. So Google's AP2 is designed to mathematically prove human intent. Aaron

13:37Unidentified speaker (UNKNOWN): Powell Let

13:38Unidentified speaker (speaker_2): me stop you right there. because I want to make sure I understand the mechanics of

13:41Unidentified speaker (UNKNOWN): this.

13:42Unidentified speaker (speaker_2): How does Google's AP2 prove human intent

13:45Unidentified speaker (UNKNOWN): if

13:46Unidentified speaker (speaker_2): the AI is acting autonomously in the moment?

13:49Unidentified speaker (speaker_3): Good question. It creates a cryptographic chain of custody.

13:52Unidentified speaker (speaker_2): Okay, how does that work?

13:54Unidentified speaker (speaker_3): When the human sets the parameters, say they tell the agent, buy server space when it dips below this price. AP2 generates an intent significance. signature.

14:03Unidentified speaker (speaker_2): Got it.

14:04Unidentified speaker (speaker_3): So when the AI actually executes the purchase hours or even days later, it carries that original human signature attached to the transaction. It's basically an audit trail baked into the protocol itself.

14:15Unidentified speaker (speaker_2): Ah, so Google is tackling the compliance and intent side.

14:18Unidentified speaker (speaker_3): Exactly.

14:19Unidentified speaker (speaker_2): Meanwhile, you've got Stripe, which basically runs Web2 payments. And they built an entirely new layer one blockchain called Tempo.

14:27Unidentified speaker (speaker_3): Yeah. Tempo is specifically optimized for high frequency AI shopping and streaming payments. Aaron

14:32Unidentified speaker (UNKNOWN): Powell

14:32Unidentified speaker (speaker_2): And they pair that with their MPP protocol to route the funds.

14:36Unidentified speaker (UNKNOWN): And

14:36Unidentified speaker (speaker_2): then of course you have the legacy giants, Visa, MasterCard, and Ant International are pushing something called the KYA framework.

14:43Unidentified speaker (speaker_3): Know your agent.

14:44Unidentified speaker (speaker_2): Right. Know your agent.

14:45Unidentified speaker (UNKNOWN): Aaron Powell

14:46Unidentified speaker (speaker_3): The legacy players know they can't rely on human KYC know your customer anymore. So they are pivoting to verify the software itself.

14:53Unidentified speaker (speaker_2): Makes sense.

14:54Unidentified speaker (speaker_3): The KYA framework ensures that if an AI agent is verified on one network, its reputation and spending history port over to others without having to re-register. They are really just trying to remain the trusted verification layer in a trustless system.

15:09Unidentified speaker (UNKNOWN): But

15:09Unidentified speaker (speaker_2): I'll push back on the idea that this is going gonna to be some massive bloodbath where one tech giant takes a all. The most surprising insight from the dossiers is that these protocols aren't necessarily competing. Aaron

15:20Unidentified speaker (UNKNOWN): Powell

15:21Unidentified speaker (speaker_3): No, they aren't. They are actually stackable.

15:23Unidentified speaker (UNKNOWN): Right. And

15:23Unidentified speaker (speaker_3): this is a crucial nuance. You don't have to choose between Google and Coinbase because they operate at different layers of the stack. They are highly composable. Aaron

15:31Unidentified speaker (UNKNOWN): Powell

15:31Unidentified speaker (speaker_2): Give me an example of how they stack. Aaron

15:33Unidentified speaker (UNKNOWN): Powell

15:33Unidentified speaker (speaker_3): Sure. Google's AP2, for example, natively integrates Coinbase's X402 as the actual crypto payment rail.

15:40Unidentified speaker (speaker_2): Oh wow.

15:41Unidentified speaker (speaker_3): Yeah. So Google handles the human intent and authorization at the application layer, and then it hands the execution off to X402, which routes the USDC stablecoin at the network layer.

15:50Unidentified speaker (speaker_2): So it's a fully integrated ecosystem. Okay. I buy the routing efficiency and I buy the stackable infrastructure. But I have to play the skeptic here for a minute.

15:59Unidentified speaker (speaker_3): Go for it.

16:00Unidentified speaker (speaker_2): Handing autonomous software real money. is terrifying.

16:04Unidentified speaker (UNKNOWN): I

16:05Unidentified speaker (speaker_2): mean, we all know LLMs hallucinate. They get confused, they make up facts, and they're definitely susceptible to manipulation.

16:11Unidentified speaker (speaker_3): Without a doubt.

16:12Unidentified speaker (speaker_2): So what stops my AI from just having a digital breakdown, blowing my entire savings account, or getting hacked by a malicious website? Aaron

16:19Unidentified speaker (UNKNOWN): Powell

16:19Unidentified speaker (speaker_3): It is arguably the most critical vulnerability in the agentic economy. And the sources we have, they don't sugarcoat the failures. We have empirical data showing this going horribly wrong. Aaron

16:29Unidentified speaker (UNKNOWN): Powell

16:29Unidentified speaker (speaker_2): Tell me about the failures.

16:30Unidentified speaker (speaker_3): Well, in May 2026, a crypto trading agent integrated with the Grok model, it was known as a banker, lost $150,000 in a matter of minutes.

16:40Unidentified speaker (speaker_2): Wow.

16:40Unidentified speaker (speaker_3): Just gone.

16:41Unidentified speaker (speaker_2): Just gone.

16:42Unidentified speaker (speaker_3): And before that, in March 2025, an agent called AIXBT lost $104,000. Aaron

16:49Unidentified speaker (UNKNOWN): Powell

16:49Unidentified speaker (speaker_2): But let's look at the mechanism. of those hacks, because it wasn't a flaw in the blockchain cryptography. The wallet itself wasn't hacked.

16:57Unidentified speaker (speaker_3): Correct. Both of those agents fell victim to prompt injection. Aaron

17:00Unidentified speaker (UNKNOWN): Powell

17:00Unidentified speaker (speaker_2): Which means it was a flaw in the AI's logic. Aaron

17:03Unidentified speaker (UNKNOWN): Powell

17:03Unidentified speaker (speaker_3): Exactly. Hackers essentially fed the AI malicious input data that tricked its reasoning layer. They convinced the AI that transferring the funds to the hacker's wallet was the logically correct execution. of its original mandate. Aaron

17:16Unidentified speaker (UNKNOWN): Powell

17:16Unidentified speaker (speaker_2): They social engineered this software.

17:18Unidentified speaker (speaker_3): They out debated the AI, basically, until it handed over the cash. Aaron

17:22Unidentified speaker (UNKNOWN): Powell

17:22Unidentified speaker (speaker_2): That is chilling. So how is the industry actually fixing this? Because if you can't trust the AI's brain, how can you possibly trust it with a wallet? Aaron

17:30Unidentified speaker (UNKNOWN): Powell

17:30Unidentified speaker (speaker_3): Well, the industry learned a really hard lesson from banker and AI XBT, which led to a fundamental design pivot.

17:37Unidentified speaker (speaker_2): Which is

17:38Unidentified speaker (speaker_3): you never put the spending rules in the AI's brain, the LLM.

17:42Unidentified speaker (UNKNOWN): Oh,

17:42Unidentified speaker (speaker_2): you separate them.

17:43Unidentified speaker (speaker_3): Yes. You hard code the rules into the wallet infrastructure itself. The

17:48Unidentified speaker (speaker_2): kill switches.

17:49Unidentified speaker (speaker_3): Exactly. They isolate the financial capability from the reasoning engine using cryptographic physical limits. So for instance, developers now use session keys. Well,

17:58Unidentified speaker (speaker_2): what does that mean in practice?

17:59Unidentified speaker (speaker_3): It means you never give the agent master control of the main wallet. You generate a temporary time-limited key that expires in, say, 10 minutes and is mathematically hardcapped at a $50 maximum spend.

18:11Unidentified speaker (speaker_2): Okay, that feels much safer. And they use allow lists too, right? Aaron

18:14Unidentified speaker (UNKNOWN): Powell

18:15Unidentified speaker (speaker_3): Yes, heavily.

18:16Unidentified speaker (UNKNOWN): The

18:16Unidentified speaker (speaker_3): agent literally cannot send money to a random hacker because the programmable wallet will only process transactions to a pre-approved, cryptographically verified list of vendors. Aaron

18:25Unidentified speaker (UNKNOWN): Powell

18:25Unidentified speaker (speaker_3): So

18:26Unidentified speaker (speaker_2): even if the LM LLM hallucinates and actively tries to send $100,000 to a malicious address,

18:31Unidentified speaker (UNKNOWN): the

18:31Unidentified speaker (speaker_2): wallet infrastructure just looks at the transaction, sees it violates the mathematical rules, and blocks it instantly. That separational logic and finance is really what makes this whole ecosystem viable. Aaron Trevor

18:46Unidentified speaker (UNKNOWN): Powell Burrus

18:46Unidentified speaker (speaker_3): It's the only way it works. Which brings us to how this actually scales in the real world over

18:50Unidentified speaker (UNKNOWN): the world.

18:51Unidentified speaker (speaker_3): next few years, particularly when it hits the regulatory wall. Aaron

18:53Unidentified speaker (UNKNOWN): Ross Powell

18:54Unidentified speaker (speaker_2): Right. Because governments aren't just going to let trillions of dollars flow invisibly between machines without having a say. Trevor

18:59Unidentified speaker (UNKNOWN): Burrus, Jr.

19:00Unidentified speaker (speaker_3): Not a chance.

19:00Unidentified speaker (speaker_2): The sources specifically highlight the American Bankers Association's guidance on the 2025 Genius Act. Aaron

19:07Unidentified speaker (UNKNOWN): Ross Powell

19:07Unidentified speaker (speaker_3): The Genius Act is essentially the regulatory framework attempting to put guardrails on this exact phenomenon. Aaron Powell

19:13Unidentified speaker (speaker_2): What are the main takeaways from it?

19:15Unidentified speaker (UNKNOWN): Aaron Powell

19:15Unidentified speaker (speaker_3): From a structural standpoint, it mandates that any stable coin used in these automated agentic transactions must be backed 100% by U.S. Treasuries in reserve. Aaron

19:24Unidentified speaker (UNKNOWN): Powell

19:24Unidentified speaker (speaker_2): Okay, so no fractional reserve banking for AI money. Aaron

19:27Unidentified speaker (UNKNOWN): Powell

19:27Unidentified speaker (speaker_3): Exactly. And it strictly prohibits yield generation on those tokens to prevent them from becoming unregistered Registered securities. But honestly, the most important piece of the legislation deals with liability. Aaron

19:37Unidentified speaker (UNKNOWN): Powell

19:38Unidentified speaker (speaker_3): Because

19:38Unidentified speaker (speaker_2): software doesn't have legal personhood. You can't put an AI agent in federal prison for money laundering. Aaron

19:44Unidentified speaker (UNKNOWN): Powell

19:44Unidentified speaker (speaker_3): Exactly. So the Genius Act firmly establishes that all KYC and anti-money laundering liability flows upward to the human or the corporation that deployed the agent. Aaron

19:54Unidentified speaker (UNKNOWN): Powell

19:55Unidentified speaker (speaker_2): The legal burden remains entirely human. Aaron

19:57Unidentified speaker (UNKNOWN): Powell

19:57Unidentified speaker (speaker_3): 100%.

19:59Unidentified speaker (speaker_2): So bringing it all together, what does this actually mean for you? listening?

20:03Unidentified speaker (UNKNOWN): If

20:03Unidentified speaker (speaker_2): you're a small business owner or an e-commerce platform or just an everyday consumer, how does this shift your reality?

20:10Unidentified speaker (speaker_3): Honestly, it means the end of rigid subscriptions and the birth of an extreme hyper-liquid pay-as-you go economy. Aaron

20:17Unidentified speaker (UNKNOWN): Powell

20:17Unidentified speaker (speaker_2): Give me an example for a small business.

20:19Unidentified speaker (speaker_3): For a small business, you won't buy monolithic software packages anymore. Your agent will dynamically route tasks and buy microservices. down to the millisecond.

20:29Unidentified speaker (UNKNOWN): And

20:29Unidentified speaker (speaker_2): for e-commerce?

20:30Unidentified speaker (speaker_3): For e-commerce, it's pretty blunt. If you don't integrate protocols like X402 or AP2, you are going to be completely invisible to the largest purchasing demographic of the next decade, which is AI agents. Aaron

20:41Unidentified speaker (UNKNOWN): Powell

20:42Unidentified speaker (speaker_2): We really are moving into an era where money moves at the speed of software, completely independent of human clicks, checkout pages, and biometric approvals.

20:50Unidentified speaker (UNKNOWN): We're

20:50Unidentified speaker (speaker_2): delegating not just our tasks, but our actual purchasing power to software that negotiates and settles in fractions of a cent on crypto rails.

20:58Unidentified speaker (speaker_3): It's a whole new world.

20:59Unidentified speaker (speaker_2): It really is.

21:01Unidentified speaker (UNKNOWN): Thank

21:01Unidentified speaker (speaker_2): you for joining us on this deep dive into the agentic economy. We've unpacked the shift from chatbots to autonomous wallets, the failure of legacy payment rails, and the massive potential of protocols like X402. X four oh two. But as we wrap up, there is one final opening. open-ended question that the data forces us to confront. Something for you to mull over as this technology matures.

21:20Unidentified speaker (speaker_3): Right. If AI agents start paying, buying, and negotiating among themselves autonomously, who will truly control the economy of the future?

21:29Unidentified speaker (speaker_2): Something to think about. Catch you on the next one.

This page was produced automatically: the audio was transcribed with nvidia/parakeet-tdt-0.6b-v3+sortformer-v2.1 and interpreted by Qwen3.6-35B-A3B-UD-IQ3_S.gguf. Transcription and interpretation can contain errors; always confirm against the original episode. Nothing here is investment advice.
All insights