People think a lot about how the iPhone changed our lives with the introduction of the app economy and how our lives revolve around this device. They think a lot less about the fact that it simultaneously killed off the dominant user interface we’d grown accustomed to for decades - the physical keyboard. I think we’re about to witness a similar paradigm shift, but this time it's the entire concept of graphical user interfaces that's on the verge of disruption.

A few months ago, I wrote about how crypto's future interface would be AI-forward - not another wallet redesign or slicker DEX, but something fundamentally different. I referenced the Rabbit R1 pin, which navigated the web on your behalf, interacting with websites and applications without traditional interfaces. While Rabbit ultimately stumbled, the point has proven directionally correct and is now materializing faster than expected via Model Context Protocol (MCP), introduced by Anthopic last year.

To grasp the importance of MCP, we first have to understand the role of an Application Programming Interface (API). Think of an API as a set of rules that allows different software applications to talk to each other. Nearly every digital service you use today - from Google Drive to your banking app - has an API. It's what allows one program to programmatically request information or trigger an action in another. Any action you take by clicking a button in an app can often be replicated by sending a structured message to its API.

Historically, APIs have been the domain of engineers. They are the tools developers use to integrate services and build new features. MCP changes this entirely. It acts as a universal translator, taking all these disparate APIs and converting them into a standardized set of "tools" that Large Language Models (LLMs) can understand and use.

When you connect your Google Drive or Gmail to tools like Claude or ChatGPT, you're using MCP. This gives the LLM context and access to your private data within those services, turning the LLM itself into your primary interface. Want to search through thousands of emails for that contract from last March? Instead of clicking through Gmail's interface, you ask Claude. Planning a dinner party? Rather than juggling between recipe sites, shopping lists, and Instacart, you describe your vision to an AI that orchestrates everything behind the scenes.

Few people realize how useful this already is. Need to do your taxes: connect Gmail ask Claude or ChatGPT to find all charitable receipts or home improvement invoices from 2025. In seconds you’ll have exactly what you need. Apps like Shortwave exemplify this power really well.

Here's where things get interesting. We're approaching a world of UI-free applications - services that exist purely as MCP connectors without traditional interfaces.

Consider the Instacart example. Today, planning a Thanksgiving dinner means juggling between recipe websites, manually calculating portions, cross-referencing dietary restrictions, building shopping lists, and navigating Instacart's interface to find each item. It's exhausting.

Now imagine simply telling your AI: "I'm hosting 20 people for Thanksgiving. Three are vegetarian, one has a nut allergy, and my budget is $300."

Your AI orchestrates a symphony of MCP-connected services, but here's the crucial part - it knows you. If you’ve used ChatGPT or Perplexity recently you may have seen how the LLM grows to understand you over time through your chat history (more on this in a bit).

It remembers you prefer rustic sunflower arrangements over formal roses from past orders. It knows your nephew's nut allergy is specifically to tree nuts, not peanuts. It recalls that Aunt Jane only drinks oat milk and your father-in-law won't eat anything with rosemary.

Instacart's MCP server handles grocery orders from multiple stores, but your AI knows to get the organic cranberries from Whole Foods because you mentioned preferring them last year. DoorDash checks restaurants for the pecan pie you don't want to bake, but automatically orders the apple alternative for your nephew. Target's MCP server sources serving platters in that earthy aesthetic you gravitate toward. Amazon finds table decorations that match the "cozy autumn" vibe from your Pinterest boards (connected via MCP, naturally). Your favorite florist receives a request for a low, sprawling centerpiece because the AI remembers you hate when flowers block conversation.

The AI doesn't just execute tasks - it applies years of learned context. It understands portion scaling from past dinners, checks ingredient conflicts across all food orders against its detailed knowledge of your guests' restrictions, optimizes budget allocation based on your typical spending patterns, and even generates a cooking timeline that accounts for the fact you always underestimate how long turkey takes. Each service remains focused on what it does best, while your AI conducts the entire experience with the accumulated wisdom of every interaction you've ever had. No clicking through categories. No re-entering the same dietary restrictions for the hundredth time. No manual cart building. Just intent to execution, refined by relationship.

It feels like science fiction - but in reality it might be what we experience by next November at the rate that models and MCP is accelerating.

The enterprise implications are already profound. A support team lead can already ask Claude: "What's causing customer frustration this month?" The AI doesn't just search tickets - it identifies patterns across thousands of interactions, correlates issues with recent deployments, generates bug reports with reproduction steps, assigns them to the right engineers based on code ownership, and even submits pull requests with proposed fixes. What once required multiple teams and systems now happens in a single conversation.

For developers, MCP represents liberation from UI/UX constraints. Instead of spending months perfecting user flows and A/B testing button placements, they can focus on what their service actually does. Build robust APIs, define clear endpoints, and let AI handle the interaction layer. A payment processor doesn't need a dashboard - just endpoints for transactions, refunds, and reporting that AI agents can orchestrate.

For users, it means the end of app fatigue. No more downloading dozens of apps, remembering different interfaces, or hunting through settings. Your AI becomes a universal remote for your digital life - one conversation to rule them all.

As an investor, the implications are really exciting. Every SaaS product, every marketplace, every service platform could become an MCP-first business. The winners won't be those with the prettiest apps or the slickest onboarding flows, but those who best expose their capabilities to AI agents. In this new paradigm, the quality of your API documentation might matter more than your user interface - because your API is your user interface.

Going back to the iPhone comment from earlier: MCP doesn't just change software - it obsoletes the entire foundation of modern computing interfaces. The smartphone, that we stare at for hours each day, is built on a fundamentally flawed premise for an AI-first world.

Think about it. Why do we need a touchscreen when our AI understands natural language? Why scroll through pages of apps when a single conversation can orchestrate dozens of services? The iPhone's grid of icons - each a portal to a different interface you must learn and navigate - looks as antiquated as a rotary phone in an MCP-enabled world.

In five years, I have a hard time believing we will be tapping glass screens the way we do today. Instead we will be getting stuff done through whatever form factor makes sense for the moment - glasses that overlay information, earpieces for private interactions, pins like that Rabbit device, or more likely something we haven't imagined yet. The device becomes secondary to the intelligence that flows through it. This shift is already occurring on desktop, where Perplexity (Comet), The Browser Company (Dia), and OpenAI (tba) are all releasing or working on AI native browsers that showcase what an agentic computing experience can and should look like with our PCs.

Hardware companies see this tsunami coming too. Apple's Vision Pro might seem like it is for watching movies in 3D. In reality it is probably their best attempt to own the next interface paradigm. OpenAI buying Jonny Ive’s IO for $1B, Meta's Ray-Bans with AI integration, Amazon's continued Alexa experiments, Google's various wearable attempts - they're all searching for the form factor that makes sense when your primary interface is conversation, not tapping.

The smartphone era is ending not because we found a better rectangle, but because we found a better way to interact with the digital world entirely. Just as the iPhone made physical keyboards seem clunky and limited, AI agents make app interfaces feel like unnecessary friction.

Of course there’s a crypto angle here too.

The crypto skeptics love to ask: "What problem is there really with credit cards for consumers?” and “Why would I give up my 2% credit card rewards for crypto payments?" Not accounting for what crypto payments offer to emerging markets in terms of access to stable currency and global digital payments, it's the wrong question for the wrong era.

First, let's address the rewards system. That 2% cash back isn't free money - it's your money, recycled through hidden fees. Every swipe costs merchants 3% plus $0.30 in interchange fees. Your local coffee shop isn't giving you rewards; they're marking up prices to cover Visa's toll. Eliminate those fees through stablecoin payments, and suddenly your $5 latte could be $4.75. Scale that across every transaction, and the savings dwarf any rewards program.

But that's still thinking in yesterday's paradigm. The real revolution isn't about replacing credit cards at the coffee shop - it's about enabling an entirely new commerce modality that credit cards can't touch.

When MCP allows AI agents execute thousands of micro-transactions on your behalf - searching paid databases, accessing premium APIs, commissioning specialized computations - the credit card model collapses. Imagine your AI assistant negotiating with a dozen different services to plan that dinner party, each charging fractions of a cent for specific data or capabilities. Traditional payment rails can't handle this volume or granularity efficiently.

This is where stablecoins connected via MCP become essential infrastructure. Your AI doesn't need a credit card number; it needs programmable money it can deploy instantly across countless services. No authorization holds, no batch processing, no $0.30 minimum fees eating up 90% of a micro-transaction's value.

Looking further ahead, MCP opens intriguing possibilities for on-chain identity and memory. Imagine your AI assistant's knowledge and context about you - your preferences, history, learned patterns - stored not in OpenAI's or Anthropic's servers, but in a self-custodial, blockchain-based memory layer. This feels critically important when you think about the power of these models and the unknowns that still exist. I love to use ChatGPT for many things but I prefer Claude for others and Gemini or Grok for still others. When a new state of the art model is released, I want to use it and I want my memories and personalization to come with me.

This sovereign AI stack would connect to LLMs via MCP while maintaining your data under your control. Switch between AI providers without losing context. Share specific memories or capabilities with services on a permissioned basis. Build a persistent, portable identity that grows more valuable over time - owned by you, not harvested by platforms.

In the context of MCP, the credit card networks may be fighting yesterday's war, optimizing for a world of human fingers tapping on screens. But when AI agents become our primary interface to the digital world, executing complex workflows across dozens of services in seconds, we need money that moves at the speed of software. When Shopify simultaneously begins exposing seller goods via ChatGPT and facilitating stablecoin payments in the same month, these things might just be connected actions.

Stablecoins aren't trying to win at credit cards' game. They're building the payment infrastructure for a game that hasn't fully started yet - one where AI orchestrates commerce at a scale and granularity that makes traditional payment methods look like dial-up internet.

The next time someone asks why they should give up their credit card rewards, remind them: crypto payments aren’t optimized for the world of apps and websites. They’re more likely for a world where AI handles the clicking, and money needs to flow as seamlessly as API calls.

The question isn't whether stablecoins can compete with credit cards for human-initiated purchases. It's whether credit cards can survive in a world where humans don't make purchases anymore - their AI agents do, thousands of times per day, for amounts too small for traditional systems to process profitably.

That world is closer than most realize. MCP is the bridge, and stablecoins are what will take us across it. As an investor exploring what’s next, I am closely watching how entrepreneurs begin to apply this new model and what new services are unlocked as a result.

People think a lot about how the iPhone changed our lives with the introduction of the app economy and how our lives revolve around this device. They think a lot less about the fact that it simultaneously killed off the dominant user interface we’d grown accustomed to for decades - the physical keyboard. I think we’re about to witness a similar paradigm shift, but this time it's the entire concept of graphical user interfaces that's on the verge of disruption.

A few months ago, I wrote about how crypto's future interface would be AI-forward - not another wallet redesign or slicker DEX, but something fundamentally different. I referenced the Rabbit R1 pin, which navigated the web on your behalf, interacting with websites and applications without traditional interfaces. While Rabbit ultimately stumbled, the point has proven directionally correct and is now materializing faster than expected via Model Context Protocol (MCP), introduced by Anthopic last year.

To grasp the importance of MCP, we first have to understand the role of an Application Programming Interface (API). Think of an API as a set of rules that allows different software applications to talk to each other. Nearly every digital service you use today - from Google Drive to your banking app - has an API. It's what allows one program to programmatically request information or trigger an action in another. Any action you take by clicking a button in an app can often be replicated by sending a structured message to its API.

Historically, APIs have been the domain of engineers. They are the tools developers use to integrate services and build new features. MCP changes this entirely. It acts as a universal translator, taking all these disparate APIs and converting them into a standardized set of "tools" that Large Language Models (LLMs) can understand and use.

When you connect your Google Drive or Gmail to tools like Claude or ChatGPT, you're using MCP. This gives the LLM context and access to your private data within those services, turning the LLM itself into your primary interface. Want to search through thousands of emails for that contract from last March? Instead of clicking through Gmail's interface, you ask Claude. Planning a dinner party? Rather than juggling between recipe sites, shopping lists, and Instacart, you describe your vision to an AI that orchestrates everything behind the scenes.

Few people realize how useful this already is. Need to do your taxes: connect Gmail ask Claude or ChatGPT to find all charitable receipts or home improvement invoices from 2025. In seconds you’ll have exactly what you need. Apps like Shortwave exemplify this power really well.

Here's where things get interesting. We're approaching a world of UI-free applications - services that exist purely as MCP connectors without traditional interfaces.

Consider the Instacart example. Today, planning a Thanksgiving dinner means juggling between recipe websites, manually calculating portions, cross-referencing dietary restrictions, building shopping lists, and navigating Instacart's interface to find each item. It's exhausting.

Now imagine simply telling your AI: "I'm hosting 20 people for Thanksgiving. Three are vegetarian, one has a nut allergy, and my budget is $300."

Your AI orchestrates a symphony of MCP-connected services, but here's the crucial part - it knows you. If you’ve used ChatGPT or Perplexity recently you may have seen how the LLM grows to understand you over time through your chat history (more on this in a bit).

It remembers you prefer rustic sunflower arrangements over formal roses from past orders. It knows your nephew's nut allergy is specifically to tree nuts, not peanuts. It recalls that Aunt Jane only drinks oat milk and your father-in-law won't eat anything with rosemary.

Instacart's MCP server handles grocery orders from multiple stores, but your AI knows to get the organic cranberries from Whole Foods because you mentioned preferring them last year. DoorDash checks restaurants for the pecan pie you don't want to bake, but automatically orders the apple alternative for your nephew. Target's MCP server sources serving platters in that earthy aesthetic you gravitate toward. Amazon finds table decorations that match the "cozy autumn" vibe from your Pinterest boards (connected via MCP, naturally). Your favorite florist receives a request for a low, sprawling centerpiece because the AI remembers you hate when flowers block conversation.

The AI doesn't just execute tasks - it applies years of learned context. It understands portion scaling from past dinners, checks ingredient conflicts across all food orders against its detailed knowledge of your guests' restrictions, optimizes budget allocation based on your typical spending patterns, and even generates a cooking timeline that accounts for the fact you always underestimate how long turkey takes. Each service remains focused on what it does best, while your AI conducts the entire experience with the accumulated wisdom of every interaction you've ever had. No clicking through categories. No re-entering the same dietary restrictions for the hundredth time. No manual cart building. Just intent to execution, refined by relationship.

It feels like science fiction - but in reality it might be what we experience by next November at the rate that models and MCP is accelerating.

The enterprise implications are already profound. A support team lead can already ask Claude: "What's causing customer frustration this month?" The AI doesn't just search tickets - it identifies patterns across thousands of interactions, correlates issues with recent deployments, generates bug reports with reproduction steps, assigns them to the right engineers based on code ownership, and even submits pull requests with proposed fixes. What once required multiple teams and systems now happens in a single conversation.

For developers, MCP represents liberation from UI/UX constraints. Instead of spending months perfecting user flows and A/B testing button placements, they can focus on what their service actually does. Build robust APIs, define clear endpoints, and let AI handle the interaction layer. A payment processor doesn't need a dashboard - just endpoints for transactions, refunds, and reporting that AI agents can orchestrate.

For users, it means the end of app fatigue. No more downloading dozens of apps, remembering different interfaces, or hunting through settings. Your AI becomes a universal remote for your digital life - one conversation to rule them all.

As an investor, the implications are really exciting. Every SaaS product, every marketplace, every service platform could become an MCP-first business. The winners won't be those with the prettiest apps or the slickest onboarding flows, but those who best expose their capabilities to AI agents. In this new paradigm, the quality of your API documentation might matter more than your user interface - because your API is your user interface.

Going back to the iPhone comment from earlier: MCP doesn't just change software - it obsoletes the entire foundation of modern computing interfaces. The smartphone, that we stare at for hours each day, is built on a fundamentally flawed premise for an AI-first world.

Think about it. Why do we need a touchscreen when our AI understands natural language? Why scroll through pages of apps when a single conversation can orchestrate dozens of services? The iPhone's grid of icons - each a portal to a different interface you must learn and navigate - looks as antiquated as a rotary phone in an MCP-enabled world.

In five years, I have a hard time believing we will be tapping glass screens the way we do today. Instead we will be getting stuff done through whatever form factor makes sense for the moment - glasses that overlay information, earpieces for private interactions, pins like that Rabbit device, or more likely something we haven't imagined yet. The device becomes secondary to the intelligence that flows through it. This shift is already occurring on desktop, where Perplexity (Comet), The Browser Company (Dia), and OpenAI (tba) are all releasing or working on AI native browsers that showcase what an agentic computing experience can and should look like with our PCs.

Hardware companies see this tsunami coming too. Apple's Vision Pro might seem like it is for watching movies in 3D. In reality it is probably their best attempt to own the next interface paradigm. OpenAI buying Jonny Ive’s IO for $1B, Meta's Ray-Bans with AI integration, Amazon's continued Alexa experiments, Google's various wearable attempts - they're all searching for the form factor that makes sense when your primary interface is conversation, not tapping.

The smartphone era is ending not because we found a better rectangle, but because we found a better way to interact with the digital world entirely. Just as the iPhone made physical keyboards seem clunky and limited, AI agents make app interfaces feel like unnecessary friction.

Of course there’s a crypto angle here too.

The crypto skeptics love to ask: "What problem is there really with credit cards for consumers?” and “Why would I give up my 2% credit card rewards for crypto payments?" Not accounting for what crypto payments offer to emerging markets in terms of access to stable currency and global digital payments, it's the wrong question for the wrong era.

First, let's address the rewards system. That 2% cash back isn't free money - it's your money, recycled through hidden fees. Every swipe costs merchants 3% plus $0.30 in interchange fees. Your local coffee shop isn't giving you rewards; they're marking up prices to cover Visa's toll. Eliminate those fees through stablecoin payments, and suddenly your $5 latte could be $4.75. Scale that across every transaction, and the savings dwarf any rewards program.

But that's still thinking in yesterday's paradigm. The real revolution isn't about replacing credit cards at the coffee shop - it's about enabling an entirely new commerce modality that credit cards can't touch.

When MCP allows AI agents execute thousands of micro-transactions on your behalf - searching paid databases, accessing premium APIs, commissioning specialized computations - the credit card model collapses. Imagine your AI assistant negotiating with a dozen different services to plan that dinner party, each charging fractions of a cent for specific data or capabilities. Traditional payment rails can't handle this volume or granularity efficiently.

This is where stablecoins connected via MCP become essential infrastructure. Your AI doesn't need a credit card number; it needs programmable money it can deploy instantly across countless services. No authorization holds, no batch processing, no $0.30 minimum fees eating up 90% of a micro-transaction's value.

Looking further ahead, MCP opens intriguing possibilities for on-chain identity and memory. Imagine your AI assistant's knowledge and context about you - your preferences, history, learned patterns - stored not in OpenAI's or Anthropic's servers, but in a self-custodial, blockchain-based memory layer. This feels critically important when you think about the power of these models and the unknowns that still exist. I love to use ChatGPT for many things but I prefer Claude for others and Gemini or Grok for still others. When a new state of the art model is released, I want to use it and I want my memories and personalization to come with me.

This sovereign AI stack would connect to LLMs via MCP while maintaining your data under your control. Switch between AI providers without losing context. Share specific memories or capabilities with services on a permissioned basis. Build a persistent, portable identity that grows more valuable over time - owned by you, not harvested by platforms.

In the context of MCP, the credit card networks may be fighting yesterday's war, optimizing for a world of human fingers tapping on screens. But when AI agents become our primary interface to the digital world, executing complex workflows across dozens of services in seconds, we need money that moves at the speed of software. When Shopify simultaneously begins exposing seller goods via ChatGPT and facilitating stablecoin payments in the same month, these things might just be connected actions.

Stablecoins aren't trying to win at credit cards' game. They're building the payment infrastructure for a game that hasn't fully started yet - one where AI orchestrates commerce at a scale and granularity that makes traditional payment methods look like dial-up internet.

The next time someone asks why they should give up their credit card rewards, remind them: crypto payments aren’t optimized for the world of apps and websites. They’re more likely for a world where AI handles the clicking, and money needs to flow as seamlessly as API calls.

The question isn't whether stablecoins can compete with credit cards for human-initiated purchases. It's whether credit cards can survive in a world where humans don't make purchases anymore - their AI agents do, thousands of times per day, for amounts too small for traditional systems to process profitably.

That world is closer than most realize. MCP is the bridge, and stablecoins are what will take us across it. As an investor exploring what’s next, I am closely watching how entrepreneurs begin to apply this new model and what new services are unlocked as a result.