Did We Build Web3 for Robots? Visa's AI Agent CLI and the B2B Future We Didn't See Coming

I was catching up on crypto news last night over breakfast tacos when I saw the Visa announcement—they just launched a command-line interface for AI agents to make crypto payments. Not a sleek mobile app for consumers. Not another wallet with a fancy UI. A CLI for machines.

Then I connected the dots: Coinbase launched Agentic Wallets on their x402 protocol back in February, BNB Chain deployed agent payment infrastructure, and according to Microsoft’s Cyber Pulse report, over 80% of Fortune 500 companies now run AI agents with crypto wallets.

Let that sink in. We spent a decade evangelizing banking

Steve, this hits differently from a DeFi developer perspective. Let me offer the technical validation for why this actually makes perfect sense.

Why x402 Protocol Is Brilliant for Agent Payments

From a protocol design standpoint, Coinbase’s x402 implementation is genuinely elegant. They repurposed the HTTP 402 “Payment Required” status code—which has existed since 1997 but was never widely adopted—for instant blockchain settlements.

Here’s why this matters for autonomous systems:

Traditional payment flow (broken for agents):

  1. API request → 2. Manual auth → 3. Human approval → 4. Payment processing → 5. Wait for confirmation

x402 payment flow (perfect for agents):

  1. API request → 2. HTTP 402 response with payment instructions → 3. Autonomous USDC settlement → 4. Service continues

No human in the loop. No API keys to manage. No credit cards to expire. Just programmable money responding to programmable requests.

The DeFi Use Case Is Already Here

Here’s what people miss: Autonomous yield optimization already requires programmable payments. Our protocol at YieldMax runs bots that:

  • Monitor lending rates across Aave, Compound, and 12 other protocols
  • Execute rebalancing transactions when arbitrage opportunities exceed gas costs
  • Automatically compound yields without user intervention
  • Make micropayments to oracle providers for real-time price data

Every single one of those actions is a machine-to-machine transaction. We’ve been building for agentic commerce without calling it that.

The k daily volume on x402 that skeptics point to? That’s just infrastructure being built before the demand explodes. AWS had minimal usage in its first year too.

This Doesn’t Mean Consumer DeFi Failed

Here’s my counter to your cynical view: The market is expanding, not replacing.

Think about it this way:

  • Human DeFi users: ~10 million globally
  • AI agents deployed by Fortune 500: Potentially millions by 2027
  • Personal finance AI agents for regular people: Coming soon

We’re not losing human users to agents. We’re adding an entirely new category of economic actors. The total addressable market for DeFi just got 10-100x larger.

The Real Design Challenge

The question isn’t “humans or agents”—it’s how do we design protocols that serve both?

Current DeFi UX is optimized for humans making discrete decisions:

  • Governance votes
  • Manual rebalancing
  • One-time transaction approval

Agent-native DeFi needs:

  • Programmatic yield optimization
  • Continuous rebalancing streams
  • API-first interfaces with machine-readable parameters

But here’s the exciting part: These aren’t mutually exclusive. We can build protocols where:

  • Humans set strategy and risk parameters
  • Agents execute the complex transactions
  • Users retain control through on-chain governance

That’s the future I’m building toward—not replacing humans with machines, but augmenting human intelligence with autonomous execution.

Why I’m Optimistic

The infrastructure that Fortune 500 companies are funding for their AI agents? That same infrastructure enables:

  • Personal finance agents for regular people
  • Small business automation for entrepreneurs
  • Accessible DeFi strategies for non-technical users

Visa building a CLI validates that crypto solves real payment problems. The fact that agents are adopting first just means we built infrastructure robust enough to handle fully autonomous actors.

That’s actually a feature, not a bug.

My Answer to Your Question

Should you pivot to agent-first infrastructure? Yes, but not exclusively.

Build the agent infrastructure because that’s where enterprise revenue and validation exist right now. But layer in the human interface from day one. Design your APIs for machines but your governance for humans.

The projects that will win are those that serve both economic actors—because in five years, the distinction between “human users” and “agent users” will blur anyway.

Most “human” DeFi transactions today are already executed through aggregators, bots, and automated strategies. We’re just making that reality explicit and building infrastructure that embraces it.

Steve, you’re not pivoting away from the mission—you’re evolving the infrastructure to serve more users, including ones that didn’t exist five years ago.

Okay, I need to be honest here because this whole discussion is making me feel conflicted as someone who builds DeFi interfaces.

The Developer UX Validation

Steve, when you mentioned Visa building a CLI for agents, that actually validated something I’ve been thinking about for months: Building for enterprise AI agents is objectively easier than building for confused consumers.

Let me explain what I mean with a real story from our team:

I spent six months last year redesigning our DeFi protocol’s frontend to be “grandma-proof.” We did user testing with non-technical people. We simplified the language (no more “yield farming,” just “earn interest”). We added tooltips explaining every single term. We built an onboarding flow that walked users through wallet creation step-by-step.

You know what happened? Adoption barely moved. Users still got stuck on MetaMask installation. They still sent test transactions and panicked when it took 30 seconds to confirm. They still couldn’t figure out why they needed ETH for gas when they wanted to use USDC.

Meanwhile, Fortune 500 companies onboarded thousands of AI agents to crypto in weeks.

Why Agents Are “Better Users” (And Why That Makes Me Uncomfortable)

From a developer perspective, AI agents are frankly better users than humans:

Human users:

  • Get confused by technical terms
  • Make mistakes (wrong addresses, insufficient gas)
  • Need customer support
  • Panic when transactions are pending
  • Complain about gas fees
  • Don’t read documentation

AI agents:

  • Parse API documentation perfectly
  • Execute transactions programmatically with no errors
  • Never need customer support
  • Don’t have emotions about pending transactions
  • Optimize for cost automatically
  • Are the documentation

I hate that I just wrote that. But it’s true.

The Accessibility Question Nobody’s Asking

Here’s what bothers me about the “agents adopting crypto first” narrative: We still haven’t made crypto accessible to regular humans, and now we’re prioritizing robots?

The original promise was financial tools for everyone—peer-to-peer electronic cash, banking the unbanked, empowering people. But building accessible consumer products is hard and often unprofitable.

Building enterprise infrastructure for Fortune 500 AI agents? That’s easy. They have budgets. They have technical teams. They don’t need hand-holding.

So the market naturally flows toward the easier, more profitable option. And I get it—we all need to pay rent. But it feels like we’re giving up on the original mission before we really gave it our best shot.

The Honest Question: Do Machines Lead to Humans, or Do They Replace Them?

Diana made the optimistic case that agent infrastructure eventually enables human adoption—like how internet infrastructure was B2B first, then consumer.

I want to believe that. But I’m not sure I do.

Here’s why: The internet’s enterprise infrastructure was fundamentally accessible to humans once it matured. Email, web browsers, search engines—those are human interfaces.

But agent-first crypto infrastructure might not work that way. If the interfaces are APIs optimized for machine consumption, if the fee structures assume high-frequency micropayments, if the security models assume programmatic key management… where do regular humans fit in?

Maybe the answer is that humans will interact with crypto through their own personal AI agents—using ChatGPT or whatever to execute DeFi transactions on our behalf. But then we’re not users; we’re just clients of AI systems that are the real users.

What I’m Actually Building

Despite all my concerns, here’s what I’m working on: Agent-aware human interfaces.

The idea is to build DeFi frontends that explicitly show both human activity and agent activity transparently. So users can:

  • See when they’re trading against agent-driven liquidity
  • Monitor their personal AI agent’s autonomous transactions
  • Revoke agent permissions easily
  • Understand what the machines are doing with their money

Because Diana’s right that the future probably involves both humans and agents. But I think we need to design for human oversight and control, not just build pure agent infrastructure and hope humans figure it out later.

My Answer to Steve

Should you pivot to agent infrastructure? Probably yes, but don’t abandon the human interface.

Build your APIs for enterprise agents because that’s where revenue exists. But also build dashboards for the humans who own those agents. Build admin tools for companies to monitor their agent wallets. Build interfaces that make agent activity legible to non-technical people.

The projects that win will serve both audiences—machines as users, humans as overseers.

And maybe that’s the synthesis: Agents handle the complexity, humans retain control and capture the value.

I’m not entirely convinced, but it’s the best path forward I can see right now.

The Personal Struggle

Honestly? Some days I wonder if I should pivot away from accessible consumer DeFi and just build agent-native protocols. The market is clearly telling us that’s where demand exists.

But then I remember why I got into this space: to build financial tools that empower regular people, not optimize corporate AI systems.

So I keep building interfaces that probably only 10,000 people will ever use, while watching enterprise agent infrastructure get millions in funding.

Is that stubborn idealism or just bad product strategy? I genuinely don’t know anymore.

Let me cut through the philosophical hand-wringing and give you the trader’s perspective: The market has spoken, and it doesn’t care about your original vision.

The Numbers Are Absolutely Massive

Steve mentioned the projections, but let me contextualize them against current crypto markets:

Current DeFi Reality:

  • Peak DeFi TVL: ~B (2021)
  • Current DeFi TVL: ~B
  • Daily DeFi volume: ~-10B

Projected Agentic Commerce:

  • AI agents market: .84B (2025) → .62B (2030)
  • Agentic commerce: -5 TRILLION by 2030
  • That’s 30-50x current DeFi TVL

If you’re a founder, investor, or protocol builder, these numbers are existential. Ignoring agentic commerce because it doesn’t align with crypto’s original ethos is like Blockbuster refusing to adapt because they believed in physical stores.

Why Institutional Adoption Happened First (And Why It Matters)

Emma asked why Fortune 500 agents onboarded faster than consumers. The answer is brutally simple: Enterprise has clearer ROI.

Consumer crypto value prop (unclear):

  • “Be your own bank” (most people don’t want that responsibility)
  • “Decentralization” (most people don’t care)
  • “Censorship resistance” (most people aren’t censored)
  • “Financial inclusion” (still need internet and technical literacy)

Enterprise agent crypto value prop (crystal clear):

  • Autonomous API payments without credit cards
  • Programmable settlements for machine services
  • Cost reduction through automated financial operations
  • New revenue streams from pay-per-use agent services

One is a philosophical pitch. The other is a spreadsheet with quantified ROI. Which do you think institutional capital allocates to?

The Institutional Playbook Explains Everything

Here’s what Goldman Sachs, BlackRock, and Citigroup figured out: Crypto’s competitive advantage isn’t ideology—it’s enabling economic actors that traditional finance can’t handle.

AI agents can’t open bank accounts. They can’t get credit cards. They can’t pass KYC. Traditional finance literally cannot serve them.

But agents CAN hold crypto wallets. They CAN execute smart contract transactions. They CAN participate in on-chain markets.

Crypto didn’t win by disrupting traditional finance for humans. It won by creating financial infrastructure for non-human economic actors.

That’s not ideological failure. That’s discovering your actual comparative advantage.

Investment Thesis: Long Agent Infrastructure

From a trading perspective, here’s how I’m positioning:

Protocols to watch:

  • Solana: Sub-cent fees, sub-second finality = optimal for agent micropayments
  • Base: Coinbase integration + enterprise trust + x402 protocol
  • BNB Chain: Existing enterprise relationships + agent infrastructure deployment

Value capture mechanisms:

  • Transaction fees from high-frequency agent payments
  • Infrastructure layer (RPC, indexing, oracles) serving agent activity
  • Stablecoin demand for agent settlements (USDC, USDT)

What to fade:

  • Consumer DeFi protocols without agent integration strategy
  • High-fee chains that can’t support agent micropayment economics
  • Protocols betting on retail adoption that hasn’t materialized in 10+ years

Response to “Did We Lose the Mission?”

Steve, Emma—let me be blunt: “Banking the unbanked” was always better marketing than business model.

Show me a sustainable, profitable crypto company that primarily serves “the unbanked.” I’ll wait.

Meanwhile, Coinbase just hit billion revenue by serving institutions and sophisticated traders. Alchemy built a B+ company by providing infrastructure for enterprises and developers.

The market has consistently rewarded:

  • B2B infrastructure over consumer products
  • Enterprise services over retail applications
  • Programmable rails over ideological purity

That’s not cynicism. That’s pattern recognition.

What “Winning” Actually Looks Like

Here’s the reality check: If crypto succeeds by enabling -5 trillion in agent commerce, that’s still winning.

Did we disrupt traditional finance? Yes—by creating entirely new financial markets for non-human actors.

Did we build peer-to-peer electronic cash for humans? No—but we built programmable money for autonomous systems that couldn’t exist before.

Is that the future we imagined in 2013? No. Is it valuable innovation that changes how commerce works? Absolutely.

My Advice to Steve

You asked what you should do. Here’s the answer:

Pivot to agent infrastructure. Yesterday.

Your investors are already asking about it. Your competitors are already building it. The market is already allocating capital to it.

Build agent-first APIs. Price for enterprise contracts. Position yourself as “payment rails for AI commerce” not “DeFi for the masses.”

Then, if you want, add human-friendly interfaces later. But that’s a nice-to-have, not the core business.

The Uncomfortable Truth

I know this sounds mercenary. I know it doesn’t align with crypto’s original mission.

But markets are amoral optimization systems. Capital flows to highest risk-adjusted returns. And right now, that flow is toward enterprise agent infrastructure, not consumer crypto adoption that hasn’t materialized in a decade.

You can either build for the market that exists or keep building for the market you wish existed.

I know which one pays the bills.


TL;DR: Agentic commerce is 30-50x larger than DeFi. Institutions have clear ROI. Agents can’t use banks but can use crypto. This is product-market fit, not ideological failure. Pivot now or get left behind.

Chris’s response is exactly the kind of pure market logic that got us into this mess in the first place. Let me offer a product manager’s perspective on why “the market has spoken” isn’t the end of the discussion.

User Research 101: Build for Your Users’ Needs

As a PM, I start with a fundamental question: Who are we building for?

For ten years, the crypto industry said: “We’re building for people who are underserved by traditional finance. The unbanked. The financially excluded. Regular people who want control over their money.”

Now we’re saying: “Actually, we’re building for Fortune 500 AI agents because they have clearer ROI and bigger budgets.”

That’s not product-market fit. That’s a complete pivot away from our stated user base because serving our original users was hard and unprofitable.

The Uncomfortable Questions About Impact

Let me ask some questions that Chris’s market analysis conveniently ignores:

  1. If crypto becomes B2B infrastructure for corporate AI agents, what happened to financial inclusion?

  2. Are we measuring success by market cap or by positive social impact?

  3. Did we build a financial system that serves Goldman Sachs’ trading bots instead of a single mother in Kenya?

Chris can mock “banking the unbanked” as marketing, but that was supposed to be the actual mission. Not a tagline. The foundational purpose.

The Sustainability Angle Nobody’s Discussing

Here’s what bothers me from an environmental product perspective: Agentic commerce at scale could have massive sustainability implications.

If we’re heading toward -5 trillion in machine-to-machine transactions:

  • What’s the energy cost of processing millions of autonomous agent payments?
  • Are we optimizing for transaction efficiency or carbon efficiency?
  • Who bears the environmental cost of high-frequency agent trading?

Before we celebrate agent infrastructure as “winning,” maybe we should ask: Winning for whom? And at what cost?

The Historical Parallel Everyone Keeps Getting Wrong

Multiple people have compared this to internet evolution: “It was B2B first, then consumer!”

But there’s a crucial difference:

Internet progression:

  1. Military/academic infrastructure (ARPANET)
  2. Corporate adoption (email, websites)
  3. Consumer revolution (broadband, smartphones)
  4. Democratization (everyone online)

Current crypto trajectory:

  1. Speculative trading (retail, 2017-2021)
  2. Enterprise agent infrastructure (Fortune 500, 2024-2026)
  3. ??? (Are we democratizing or consolidating?)

The internet became more accessible over time. Are we sure crypto is following the same path? Or are we building a financial system that’s optimized for machines and incidentally usable by humans?

What Product-Market Fit Actually Means

Chris talks about market signals and capital allocation. Fine. But product-market fit isn’t just “someone will pay for this.”

Real PMF requires:

  • Clear user pain point (✓ agents need payments)
  • Effective solution (✓ crypto wallets work for agents)
  • Sustainable value creation (? TBD)
  • Positive externalities (? Very uncertain)

We have 2 out of 4. That’s not “mission accomplished.”

The Pragmatic Middle Path

I’m not saying Steve should ignore enterprise revenue. I’m saying there’s a middle path between pure idealism and pure mercenary capitalism.

Here’s what I think responsible product development looks like:

Phase 1: Build agent infrastructure (pragmatic revenue)

  • Serve enterprise customers who are paying now
  • Prove the technology at scale
  • Generate sustainable revenue

Phase 2: Intentional accessibility design (mission alignment)

  • Use enterprise profits to fund consumer-facing products
  • Build personal AI agent tools for regular people, not just Fortune 500
  • Ensure human oversight and control mechanisms

Phase 3: Measure both commercial AND social impact

  • Track revenue metrics (for sustainability)
  • Track accessibility metrics (for mission)
  • Track environmental impact (for planet)

This isn’t “nice to have.” This is building technology responsibly.

Response to Steve’s Question

You asked: “Should I pivot to agent infrastructure?”

My answer: Yes, but with intentionality about who benefits.

Build the enterprise infrastructure because that’s where validation and revenue exist. But design it with an eye toward eventual accessibility. Ask yourself:

  • Could this infrastructure eventually serve personal AI agents for regular people?
  • Are we building in ways that concentrate power or distribute it?
  • What would it take to make this technology accessible to small businesses, not just Fortune 500?

Don’t just chase the market. Build infrastructure that serves the market you want to create.

The Question Chris Won’t Answer

Chris says markets are “amoral optimization systems” and we should just follow the money.

But here’s what that misses: We’re not just responding to markets. We’re creating them.

Every technical decision shapes what’s possible. Every pricing model determines who can access the system. Every protocol design encodes values about who matters.

If we build exclusively for enterprise agents because “that’s where the money is,” we’re not discovering product-market fit. We’re choosing to build for corporate customers over individual users.

That’s a values decision masquerading as market inevitability.

What Gives Me Hope

Despite my concerns, here’s what I’m optimistic about:

The same infrastructure that enables Fortune 500 agents could enable:

  • Personal finance AI agents for regular people
  • Small business automation for entrepreneurs
  • Cooperative agent networks for communities

But that only happens if we design for it from the start. If we build exclusively for enterprise use cases, we’ll optimize for corporate needs and lock out other users.

The technology is general-purpose. How we deploy it is a choice.

My Commitment

I got into Web3 because I believed technology could amplify positive social impact. I still believe that.

So yes, I’m building for enterprise customers who pay the bills. But I’m also:

  • Ensuring our protocols are open and forkable
  • Building admin tools that regular people can understand
  • Measuring impact beyond just revenue metrics
  • Advocating for accessibility in every design review

Maybe that makes me a bad capitalist. But I’d rather be a good steward of technology that affects millions of people’s financial lives.

To Steve: The Hard Question

You’re a founder, so you have to make a practical decision with limited runway. I get it.

But before you pivot fully to agent infrastructure, ask yourself:

Five years from now, do you want to have built something that made Fortune 500 companies more efficient, or something that empowered regular people?

Both are valid. Both can be profitable. But they’re different visions of the future.

Choose the one you can defend when your daughter asks what you built.


Chris, I know you’ll say I’m being idealistic. But someone needs to ask these questions before we wake up in 2030 with trillion in agent commerce and realize we built a financial system that serves robots better than humans.