|
GM. This is Milk Road PRO, cutting through the agentic economy hype to find where crypto actually wins. |
Todayβs edition features our latest PRO report digging into the agent economy: why itβs already here and how it could reshape crypto. Youβll get the opening below, with the rest on the site. |
Hereβs what we cover in this report: |
π€ Why agents are already driving AI growth (and what that means). βοΈ Where blockchains have a real edge (and where they donβt). π How this actually impacts tokens, portfolios, and the next few years.
|
Consensus Miami is one of the largest digital asset conferences thatβs going all in on crypto and agentic commerce. Grab your passes at 20% off. |
|
|
BLOCKCHAINS WIN. TOKENS MAYBE. |
Anthropic and OpenAI are growing faster than human usage alone can explain. |
The reason: agents consume vastly more tokens per task than humans do in a chat window. |
They run in loops, call tools, and work for hours without supervision. |
π The agentic economy isn't a far-future bet. It's already here. |
It clicked for me back in January. |
I got so excited about an AI tutorial that I sent it to my wife while she was in the maternity ward, giving birth to our second kid. Not my finest βhusband momentβ. π€ |
But that was when I stopped playing with AI tools and actually started building my life around them. Once something is working for you 24/7, you stop wondering if this is real. |
So what does it mean for crypto? |
My thesis: Agent activity is going to explode. |
The real question is who is going to benefit from it the most. |
There are genuine tailwinds pulling it onto public blockchains, and genuine headwinds that could push it onto permissioned, bank-operated rails instead. |
What we're doing today is taking an honest look at both. |
|
WHAT THE AGENT ECONOMY ACTUALLY IS |
A quick framing before we go on. |
An agent isn't a chatbot. |
A chatbot answers a question and waits. An agent takes a goal, runs in a loop, calls tools, and makes decisions without step-by-step supervision. The human shifts from operator to principal. |
π You're no longer pressing the buttons. You're setting the direction and checking the result. |
That shift is the whole thing. Once software can pursue goals autonomously, it stops being a tool and starts behaving like an economic actor. It consumes resources. It makes decisions with costs. It produces output with value. It can pay for what it uses and charge for what it produces. |
None of that was true of the chatbot generation we were all using 18 months ago. |
The skeptical response is that this is mostly demos. Fair. But the aggregate usage data tells a different story. |
|
AI token spend is up 13x since Jan, 2025. Humans have likely fueled the majority of this growth... but there's a limit to what humans can do. A person will only have so many questions in a day - making human chat sessions bounded. |
Agentic tasks? Not so much. A coding agent refactoring a codebase, a research agent compiling a report, a customer-service agent handling multi-turn resolution: each consumes orders of magnitude more tokens than a human chat. |
The numbers match the experience. The agent economy isn't a future projection. It's what's driving the curves right now. |
π Agents are a new category of customer. |
But the financial system we built over the last century wasn't built for them - so where do they fit? That's the question the rest of this report is looking to answer. |
WHY CURRENT FINANCIAL RAILS BREAK FOR AGENTS |
Ok, but who cares what kind of money is used, and which rails it is sent on? |
Crypto, debit, credit card... they all work, right? |
Why is crypto such a big part of the 'agential payments' conversation? |
Well, traditional rails have five sticking points when it comes to agential commerce, and they all trace back to the same root: the infrastructure was designed around human identity, human schedules, and human error rates: |
Identity: An agent has no passport. The account model assumes one human per account. Agents spin up and shut down in seconds, scale horizontally, and have no bank branch to walk into. Human KYC just wasn't built for them. Time: Agents run continuously. Traditional rails have settlement windows, cutoff times, and banking calendars. A cross-border payment initiated on Saturday night waits until Monday. Disputes: Card networks work because humans catch errors and call the bank. An agent running thousands of transactions a day cannot. You either rebuild dispute resolution for machines (nobody has so far) or you need rails where settlement is final by default. Traditional rails are deliberately reversible. For humans, thatβs a feature - but for agents? Itβs liability. Trust: The financial system runs on institutions vouching for each other. Agents share no institutional chain. The only verification two agents can run is cryptographic. Speed: An agent paying for an API call in real time cannot wait three business days, and sometimes cannot wait three seconds. It needs payment cleared and task resumed to happen in the same moment.
|
Together, these aren't five separate problems a bank can patch. They're one problem: the design assumptions of the infrastructure are wrong for the customer. |
π Public blockchains weren't built for the agent economy. |
They were built for a different ideological project entirely. But their design choices, pseudonymous accounts anyone can create, 24/7 deterministic settlement, cryptographic rather than institutional trust, programmable constraints, composable protocols, all happen to describe what an agent economy would need if you designed one from scratch. |
The AI and crypto stacks developed on completely separate trajectories. They just need the same infrastructure. |
That structural fit is the bull case. |
|
| Where do you go to hear people talk about crypto, AI, and real capital? | If you want surface-level takes, then itβs probably in my group chat. | But if you want institutions with deep pockets, youβll have to head to Consensus Miami. | Consensus Miami is one of the largest digital asset conferences thatβs going all in on crypto and agentic commerce. | Here are the key details: | | The best part? | You can get an exclusive 20% discount on passes with code MILKROAD. | Grab your tickets at 20% off. |
|
|
|
THE TAILWINDS |
We showed why traditional rails break. |
This section shows what blockchains get right and it's not an accident. |
Three tailwinds matter. They stack. |
1. Agents can finally hold money |
Blockchains are the first financial infrastructure where software can natively hold, receive, and send value without an institutional intermediary. |
An agent can have a wallet. That wallet is the agent's, not its principal's. |
It can receive payments, make payments, hold balances, and transact with wallets owned by other agents. No bank account required. |
This sounds small. It isn't. |
Without it, agents can only spend through their principal's accounts. Every transaction bottlenecks through human KYC, human identity, a human banking relationship. |
That's fine for demos. It's fatal for machine-to-machine commerce at scale. |
π Blockchains dissolve the bottleneck. |
An agent is spun up with an address in seconds, funded programmatically, and settles with counterparties it will never meet without asking permission from any intermediary. |
This isn't a marginal improvement. |
It's the difference between the agent economy existing and not existing. |
2. Principals can delegate safely |
The scariest thing about giving an agent money is that it might spend it wrong: |
It might misunderstand the task. It might get hijacked by a prompt injection. It might transact with a malicious counterparty.
|
Traditional finance handles this poorly. |
You can set a card limit or freeze an account. |
You cannot natively express: "spend up to $500 per month, only on compute from these providers, only in transactions under $50, hard stop at $100 daily, everything else auto-rejected and logged." |
That's not a card setting. That's a smart contract. |
Account abstraction, spending policies, session keys, and multisig let a principal grant an agent cryptographically scoped authority. |
The agent doesn't ask permission. The rails refuse to let it act outside policy. |
If it's compromised, damage is bounded by the contract, not by how fast anyone notices. |
My view: this is a precondition for serious capital delegation to agents. No one sane gives an autonomous system real money without constraints enforced in code. |
3. Markets reorganize around merit |
Agents have no brand loyalty. No relationships. No nostalgia for their parents' bank. |
They evaluate hundreds of providers in parallel and route to whatever wins on measurable criteria: price, speed, reliability, security, uptime. |
The behavioral moats protecting incumbents: |
Habit. Familiarity. Marketing. Relationship selling.
|
These don't apply to a customer who is software. |
One caveat worth naming: brand doesn't die, it migrates. |
Whoever owns the agent interface owns the routing decision. |
Your bank's agent will favor your bank. |
The "perfectly neutral optimizer" is an abstraction, but not usually the deployed reality. |
But the direction is clear. |
Measurable performance matters more. API quality, liquidity depth, and transparent track records become what providers compete on. |
This favors crypto. |
β A human might still choose the bank because of the relationship. |
β
The agent has no relationship to override the math. |
Agents need money they can control. Principals need to delegate safely. Once both are in place, capital routes are on merit. |
That's the bull case. |
The next section is where we make the strongest case against it. |
THE HEADWINDS |
Four forces could route the agent economy away from public blockchains entirely. We take them seriously, in the order they carry weight. |
1. Traditional players build their own rails |
The single strongest argument against this thesis is that incumbents aren't standing still. |
Visa, Mastercard, Stripe, the major banks, and the large fintechs are building agent-capable infrastructure on permissioned stablecoins and closed networks. |
Enterprise buyers aren't crypto-native. |
They're procurement teams, legal departments, and CFOs. |
They want an SLA, an account manager, and someone to sue if things break. |
A bank-operated stablecoin on a permissioned network offers all of that. |
Public blockchains mostly don't. |
For a CFO deciding how agents spend company money, the institutional offering is the path of least resistance, even when it's technically inferior. |
Path of least resistance tends to win large markets. |
π This is the scenario where our thesis is right about what agents need and wrong about which rails end up serving them. |
2. Capture at the agent layer |
Suppose crypto wins the rails fight. Agents transact on public blockchains. The infrastructure thesis works. |
The agents themselves will still be operated by a very small number of companies. |
The frontier model market is already concentrated. |
Agent frameworks consolidate around a handful of providers. |
The consumer surface will probably be dominated by two or three firms. The same ones that dominate consumer AI today. |
Whoever owns the agent layer owns the routing decision. |
Defaults, partnerships, whitelists, revenue-share deals with specific stablecoin issuers: all of it reintroduces exactly the intermediation the brand-agnostic tailwind was supposed to eliminate. |
Crypto wins the technical argument and loses the economic one. Public blockchains become plumbing. Value accrues upstack. |
This is roughly what happened to the open internet. The protocols are free. The value went to the platforms. |
3. The regulatory fog |
Autonomous agents transacting with money raise questions that don't have answers yet: |
Who is liable when an agent makes a bad trade? If an agent signs a malicious smart contract, is that the user, the agent, or the developerβs fault? Does an agent need its own KYC? Which jurisdiction's securities laws apply when an agent in one country buys an onchain asset issued in another?
|
These aren't philosophical questions. |
They're the exact questions institutional counsel asks before approving an agent to move company capital. Until answers exist, firms with the most to deploy will deploy slowly or not at all. |
The asymmetry matters. |
Banks know how to work with regulators. They've been doing it for a century. |
When a regulator wants to talk to Ethereum, no one picks up the phone. |
That's a feature of decentralization and a bug for institutional adoption. |
4. The demo-to-reality gap |
A lot of what gets called "the agent economy" today isn't the agent economy. It's demos, experiments, and narrative-driven launches. Real recurring machine-to-machine commerce is still a rounding error against global payment volumes. |
Enterprise agent adoption is slower than the Twitter timeline suggests. |
Security concerns are legitimate. |
The tasks agents can complete end-to-end, without human supervision, are narrower than marketing implies. The gap between "agent can do this in a demo" and "agent can be trusted unsupervised with big money" is wider than most admit. |
If that gap closes slowly, crypto's structural advantages stay true in principle but matter at scale later: 2028 or 2030 instead of 2026. |
My honest read: Iβm confident about the direction and less confident about the pace. |
In technology, being right about direction and wrong about timing is a familiar way to lose money. |
WHERE THE FIGHT ACTUALLY GETS DECIDED |
Three battles will decide the outcome. Each is unsettled. Each will resolve within 18-24 months. Watch these, not the headlines: |
1. Agent-to-agent payments |
When two agents owned by different principals exchange value, what rails do they use? This is the single most consequential fight. |
Crypto wins when: major AI labs ship SDKs with first-class public stablecoin support. Developer defaults compound. By late 2027, "my agent paid another agent" means a stablecoin on a public chain by default. |
Crypto loses when: Visa, Mastercard, and bank consortiums ship agent-capable APIs with compliance baked in. Frontier labs integrate them because enterprise customers demand it. Permissioned stablecoins become the default inside these systems. |
π Signal to watch: which stablecoins get integrated into the SDKs of the major AI labs. Developer defaults compound faster than most people expect. |
THIS IS A BIG REPORT π | As we mentioned at the beginning, these emails can only hold so much content so to read the rest of this PRO report, youβll need to click here and read it on our website. | |
|
|
|
BITE-SIZED COOKIES FOR THE ROAD πͺ |
The entire industry is headed to Miami. Get 20% off Consensus Miami passes with code MILKROAD.* |
Good lord! Michael Saylor says $10M per BTC is "the endgame." That's a 130x from here. |
We love to see it: Meta is paying creators in USDC on Solana and Polygon. Colombia and the Philippines first. 160+ countries by year-end. |
Macro view: OPEC breaks up, the Fed chair changes, and earnings season shakes up markets. |
Higher earnings rates. Lower borrowing costs. Up to 0.5% cashback. New U.S. clients get these benefits on Nexo for 30 days.** |
*this is sponsored content. **this is partner content. |
|
|
Grab your tickets at 20% off. |
| | RATE TODAYβS EDITION | What'd you think of today's edition? | |
|
|
|
MILKY MEMES π€£ |
|
|
|
ROADIE REVIEW OF THE DAY π₯ |
|
VITALIK PIC OF THE DAY |
|
|