Anti Crypto Crypto Talk

A few weeks ago, Peter Steinberger (the guy who built PSPDFKit into a market leader, raised a €100M+ strategic investment from Insight Partners, then built OpenClaw, one of the most popular open-source AI agent systems) was asked on X: “What’s the best advice for a guy in his 20s?”

His answer, viewed almost 900,000 times: “Don’t waste time with crypto.”

At first I was shocked. Then annoyed. Then, uncomfortably, I realized I could partly agree.

What happened next proved his point about crypto’s surface layer. Scammers snatched his old accounts during an Anthropic-forced rename (from “Clawd” to “Moltbot,” too close to “Claude”), launched a fake $CLAWD token on Solana that hit $16 million in market cap, and harassed him until he had to publicly say: “Crypto folks, stop harassing me.”

That’s the experience. Not an edge case. The experience.

But I had to sit with his tweet, because I spent my 20s with crypto. The conclusion I came to was more nuanced than 280 characters allow. If you spent your 20s trading crypto, yes, you probably wasted your time. If you spent your 20s building in crypto, you learned distributed systems, cryptography, incentive design, and how to ship products to zero users for years. That’s not wasted time. That’s a specific kind of training that very few people have.

This essay is my attempt to figure out which one I did.

Over time I started noticing a pattern: every crypto cycle seemed to contain a real technical breakthrough, but aimed at the wrong user. The infrastructure always worked. The protocols never crashed. People did. And only now, with AI agents needing to transact billions of times a day, does the intended user seem to have finally arrived.

Crypto before me

Crypto started way before my journey in it. The ideas (cryptography for individual privacy, digital cash, peer-to-peer networks) go back to the late 1980s.

In 1988, Timothy C. May distributed “The Crypto Anarchist Manifesto” at the Crypto ‘88 conference. Its opening line deliberately echoed Marx: “A specter is haunting the modern world, the specter of crypto anarchy.” May predicted that computer technology would provide “the ability for individuals and groups to communicate and interact with each other in a totally anonymous manner” and that this would “alter completely the nature of government regulation, the ability to tax and control economic interactions.”

In 1993, Eric Hughes published “A Cypherpunk’s Manifesto,” which laid out the philosophy more precisely. Its most quoted line: “Privacy is necessary for an open society in the electronic age.” But the line that matters most for understanding what came after is this one: “Cypherpunks write code.” Not manifestos. Not tweets. Code.

David Chaum, often called the godfather of digital cash, had been working on this since the early 1980s. His 1982 paper on blind signatures enabled anonymous electronic payments, and in 1989 he founded DigiCash. The product, eCash, was piloted with Mark Twain Bank in Missouri in 1995. Microsoft reportedly offered around $100 million to integrate eCash into Windows 95. Chaum rejected it; he wanted more.

DigiCash failed for reasons that would become painfully familiar to anyone building in crypto over the next three decades. The chicken-and-egg adoption problem: users complained there weren’t enough merchants, merchants said there weren’t enough users. The technology was ahead of the market. The internet wasn’t mainstream yet. And Chaum, by multiple accounts, was difficult to work with. DigiCash declared bankruptcy in 1998.

Hal Finney, an early PGP developer and cypherpunks mailing list regular, built Reusable Proofs of Work (RPOW) in 2004, a direct precursor to Bitcoin’s proof-of-work system. On January 10, 2009, he tweeted “Running bitcoin.” Two days later, on January 12, he received 10 BTC from Satoshi Nakamoto in block 170, the first person-to-person Bitcoin transaction in history.

The first implementation of a cryptocurrency, money in peer-to-peer networks using hard cryptography, was Bitcoin. Satoshi Nakamoto published the whitepaper in October 2008, and the network went live in January 2009. It solved the double-spending problem without a central authority, using a blockchain as a public ledger and proof-of-work as the consensus mechanism. For the first time, digital scarcity existed.

For several years, Bitcoin was mostly an experiment among cryptographers and libertarians. But slowly, it started attracting attention from people who saw it as either digital gold, a political statement, or a speculative opportunity. Often all three at once.

Discovering Ethereum

Like many of you, I was a broke college student in Lisbon. When I was working on my master thesis at Técnico I had lots of time but little money. One day one of my good friends joined me for a beer at Arco do Cego. He mentioned he’d made an investment and was up 3x already in a few weeks. At first that was captivating, but then he added that the technology behind it could’ve been an interesting thing for me and that I should take a look.

That led me to the Ethereum whitepaper. Vitalik Buterin had published it in late November 2013, when he was 19 years old. The Ethereum crowdsale ran from July to September 2014, raising about 31,500 BTC, roughly $18 million at the time. The network launched on July 30, 2015.

By the time I was reading about it, Ethereum had already survived its first existential crisis: the DAO hack of June 2016. The DAO was a decentralized investment fund that raised $150 million, the largest crowdfunding campaign in history at the time. An attacker exploited a reentrancy vulnerability in the smart contract and drained 3.6 million ETH, worth about $60 million. The community voted to hard fork and reverse the theft. Those who disagreed, believing “code is law” should be absolute, continued the original chain as Ethereum Classic.

I started researching online, first on Reddit and then on Twitter (now X). Something both lucrative and technically fascinating was happening.

After reading the whitepaper, one core idea struck me. Buterin described Ethereum as “a blockchain with a built-in Turing-complete programming language”, essentially a platform where anyone could write and deploy programs that would run on a decentralized network. My interpretation was simpler: this was a cloud, but the things that happen on it are recorded forever. The logic you upload is public and cannot be changed, which builds trust.

I became interested in the concept of unstoppable computing. As a broke student, paying for cloud infrastructure wasn’t an obvious choice. At the time, the cloud was still relatively new and not cheap. But with Ethereum, I could pay once to deploy my app and then users, when they wanted to use it, would pay for the compute. That idea was amazing.

The first thing that excited me about Ethereum wasn’t making money. It was free infrastructure. And yet the entire industry spent the next eight years optimizing for financial outcomes instead of product ones. My whole story is the arc of that divergence.

Naively, I thought this model was amazing. The industry as a whole shared that naivety for the next seven years. It was considered a feature until about 2023, not a bug.

Here’s what it means in practice: to interact with any application on Ethereum, users have to acquire a native currency called gas. Gas is a translation of how much it costs to execute and include a request on the network. Every click, every action, costs real money. Imagine paying every time you open a web page in your browser. That’s what blockchains were asking users to do.

The principle is right foundationally: someone has to pay for computation, and if that computation is decentralized, it makes sense to price it. But this was a massive User Experience problem for consumer apps. Crypto demanded expertise as the entry fee. You needed to understand wallets, private keys, gas limits, gas prices, and network congestion before you could use even the simplest application.

It would take years, and a completely different type of user, a non-human one, before this model actually made sense.

Crypto was mostly Bitcoin and Ethereum. Bitcoin as a currency and store of value (some people called it digital gold). Ethereum was the future app store, at least that was its promise. I was always more interested in the building side, and although you can arguably build on Bitcoin, Ethereum was designed from scratch as a developer platform.

The financialization trap

In the Ethereum space, the most exciting thing happening in 2017 were token sales in the form of ICOs, Initial Coin Offerings. Projects would raise funds in ETH and lock token distribution in a smart contract using a standard called ERC20. An ERC20 token could be written in about 50 lines of Solidity code.

Personally, this wasn’t very exciting to me. I was too naive. Had I not been, maybe I could be sitting on a beach somewhere drinking margaritas. ICOs were a very effective way to capitalize early, but most projects ended up dying.

The numbers are brutal. According to a Satis Group study, roughly 80% of ICOs in 2017 were identified as scams. Of the tracked projects that collectively raised billions, the vast majority of tokens were worth less than their ICO price within a year.

Some were outright frauds. Bitconnect, immortalized by the “hey hey hey” meme, was a textbook Ponzi scheme that collapsed spectacularly in January 2018. OneCoin defrauded investors of an estimated $4 billion; its founder Ruja Ignatova remains on the FBI’s most wanted list.

Some raised hundreds of millions and delivered almost nothing. EOS raised $4 billion, the largest ICO ever, and is effectively irrelevant today. Tezos raised $232 million and spent years in legal battles before producing a marginally used network.

And some survived. Chainlink (LINK) is now core infrastructure for connecting blockchains to real-world data. Binance (BNB) became the largest crypto exchange in the world. Aave started as ETHLend during the ICO era and became one of the most important DeFi protocols.

The ratio of dead to alive from that era is probably 95 to 5. But naivety cut both ways: I missed the upside of the survivors, and I also missed the downside of the scams. That might have been the better trade.

ICOs were the first time crypto discovered that its most effective product was fundraising itself. Not apps. Not decentralization. Just a new way to raise money with less friction and fewer gatekeepers. The technology worked perfectly; it did exactly what it was designed to do. The problem was that what it was designed to do wasn’t building products. It was extracting capital.

Around late 2017, a new thing popped up: CryptoKitties, built by a Vancouver team called Axiom Zen (later Dapper Labs). It was an app on Ethereum: you could buy a digital cat, breed it with others, and create more cats. This was the first successful example of a game that demonstrated digital ownership.

CryptoKitties proposed a new token standard, ERC721, for digital collectibles. The key difference from ERC20: tokens are non-fungible. Each cat was unique and couldn’t be replicated. One CryptoKitty is not interchangeable with another. This was the birth of NFTs.

The impact on Ethereum was immediate and dramatic. At its peak in early December 2017, CryptoKitties accounted for roughly 25% of all Ethereum network traffic. Gas prices spiked to the point where other applications became nearly unusable. Ethereum miners had to increase the gas limit to handle the load. The most expensive CryptoKitty, “Genesis” (Kitty #1), sold for about 246.9 ETH, roughly $117,000 at the time.

CryptoKitties was, in retrospect, a stress test. It proved both the concept (digital scarcity works, people will pay for it) and the limitation (Ethereum couldn’t scale). This single game became a catalyst for years of scalability research: Layer 2s, sharding, and eventually the move to Proof of Stake.

OpenSea was one of the first platforms to understand the potential of this space. Founded in late 2017 by Devin Finzer and Alex Atallah, it built a platform and indexer of NFTs. Every NFT contract deployed on Ethereum would get captured by OpenSea; they’d index the metadata and provide a marketplace for buying and selling.

The major problem people identified early was that metadata, the actual images and files, were stored outside the blockchain, usually on centralized servers. If those servers went down, the NFT became an empty record of ownership pointing to nothing. You owned a receipt for a file that no longer existed.

People outside crypto complained that NFTs were just images you could right-click save. I dismissed this as reductive, but the reality is that beyond financialization, ownership was never used for anything else. Which made it, honestly, a fair point.

There’s a connection worth drawing here that goes beyond analogy. OpenSea’s CTO and co-founder, Alex Atallah, left in July 2022 after building the dominant NFT marketplace. His first experiment after leaving was window.ai, essentially MetaMask for LLMs. It was a browser extension where you’d add your API keys and websites could use the injected methods for inference. Same pattern as MetaMask injecting window.ethereum, now it was window.ai.

Window.ai didn’t take off, for the same reason MetaMask was always clunky for normal users: too much friction. But then Atallah pivoted to OpenRouter, which became exactly what OpenSea was for NFTs: an aggregator and indexer, the default interface for a new primitive. OpenSea indexed NFT contracts. OpenRouter indexes AI models. Both captured their market by building the marketplace layer on top of open standards. The person who builds the marketplace almost always captures more value than what gets listed on it.

What’s remarkable is that it’s not just a parallel; it’s the same person applying the same playbook to a different primitive. OpenRouter raised $40 million and scaled to over $100 million in annualized inference spend. Some builders ride these waves with the same strategy, and it works every time.

The crypto market reached all-time highs in late 2017. Bitcoin hit approximately $20,000 in December. By January 7, 2018, the total crypto market cap peaked at roughly $830 billion. And then it crashed.

The crash was driven by a combination of forces. ICO treasuries that had raised in ETH were liquidating to fund operations, creating massive sell pressure. Retail investors who bought at the top were panic selling. Regulators, particularly the SEC, started investigating ICOs and signaling enforcement actions. The speculative fever broke, and there was nothing underneath to support the price.

By December 2018, the total crypto market cap had fallen to roughly $128 billion, an 85% decline from the peak. Bitcoin fell below $4,000. ETH dropped from over $1,400 to under $100.

For almost two years, the crypto community lived in disbelief. Some folks still believed in the vision. Others left the space entirely. Most teams that were true to the ethos and financially well-positioned kept building.

Innovation went underground. There were barely any users. But below the surface, something important was happening.

A few new ideas were circulating that some people were calling DeFi, decentralized finance. The goal was building financial use cases on open networks. Uniswap, created by Hayden Adams (who had been laid off from his engineering job and taught himself Ethereum development), launched in November 2018. It was a decentralized exchange, but unlike a traditional exchange that matches buyers and sellers through an order book, Uniswap used an automated market maker. Liquidity providers deposit pairs of tokens into pools. Anyone can trade against these pools, and a mathematical formula automatically sets the price based on the ratio of assets. No order book, no intermediary, no company that can freeze your funds. Just code.

Aave was building lending primitives. Compound was creating algorithmic interest rate protocols. MakerDAO had launched DAI, a decentralized stablecoin. Ethereum was working on Proof of Stake, which had been in the roadmap since 2014.

None of this was exciting. None of it made headlines. But it was the infrastructure layer that everything in 2020 and 2021 would run on.

The uncomfortable truth about crypto: most of the actual innovation happens when nobody is paying attention. The hype cycles don’t produce the innovation; they consume it.

Covid hit the world. But before DeFi’s summer, there was DeFi’s near-death experience.

On March 12, 2020, “Black Thursday,” ETH crashed 43% in a single day. Bitcoin fell from about $7,900 to $4,800. The crash was triggered by the global pandemic panic and cascading liquidations across crypto.

MakerDAO was hit particularly hard. Ethereum’s network became so congested that liquidation bots couldn’t process transactions. Some liquidators won Maker auctions for literally 0 DAI, acquiring ETH collateral for free because competing bids couldn’t get through the network. MakerDAO lost approximately $8.3 million in undercollateralized debt and had to hold an emergency MKR token auction to recapitalize.

Black Thursday nearly killed DeFi. Instead, it stress-tested it and the protocols survived, barely.

Around the summer, DeFi started growing. The catalyst was Compound.

On June 15, 2020, Compound began distributing COMP governance tokens to lenders and borrowers based on their protocol usage. COMP surged to approximately $330 within days of launching, making Compound a billion-dollar protocol overnight.

The mechanic was recursive and addictive: lend assets, earn COMP, borrow against your position, lend the borrowed assets, earn more COMP, repeat. Effective yields sometimes exceeded 100% APY. Every protocol rushed to copy it. Balancer, Yearn Finance (with its “fair launch” of YFI in July), and dozens of others followed within weeks.

Then came the vampire attack. SushiSwap forked Uniswap’s code, added a SUSHI token, and tried to drain Uniswap’s liquidity. It was brazen, a copy-paste with a financial incentive bolted on. It worked temporarily. And it forced Uniswap’s hand.

In September 2020, Uniswap did its own token airdrop, 400 UNI to every wallet that had ever used the protocol. At launch, those 400 tokens were worth about $1,200. A few months later, almost $20,000. It rewarded early users retroactively and kicked off an entire meta-game of using protocols early to qualify for future airdrops.

The total value locked in DeFi went from about $680 million in January 2020 to over $15 billion by December. Under $1 billion to over $15 billion in six months.

And then: Sushi, farms of all types of fruits and vegetables, farming yields of 10,000%. The hyper-financialization of everything. Farms of farms of farms. Every week a new protocol, a new fork, a new token. Most of them copied Uniswap’s code, added a token, and called it innovation.

I didn’t understand any of this. I thought everyone was a genius and all I did was simple lending of USDC in some protocols that gave me 10% yield.

In hindsight, the boring simple use case, earn yield on stablecoins, was the real product. Everything built on top was leverage and reflexivity dressed up as innovation. The people who treated DeFi as a savings account did fine. The people chasing 10,000% were mostly providing exit liquidity for earlier farmers. The “geniuses” were the ones who got in first and got out in time.

DeFi Summer was the first time crypto products had real, measurable utility. You could lend, borrow, exchange, financial primitives that people actually used. But the moment it worked, the industry turned it into a casino. The progression from Uniswap (useful) to SushiSwap (copy with token incentive) to random food farms (pure ponzinomics) happened in weeks, not years.

DeFi kept growing. And then came NFT Summer. This time it wasn’t cats.

This was partly fueled by well-capitalized users from early DeFi cycles, partly by Covid-era boredom and stimulus money, and partly by a single moment that changed everything.

In March 2021, a digital artist called Beeple sold an NFT titled “Everydays: The First 5000 Days” at Christie’s for $69.3 million. That single sale brought more attention to crypto than years of DeFi innovation combined. Suddenly every artist, celebrity, and brand wanted in.

PFP projects, profile picture collections, exploded. Bored Ape Yacht Club launched in April 2021. A single mint cost 0.08 ETH (around $200). Months later, the floor price was six figures. People were putting apes on their Twitter profiles, getting tattoos of them, building their entire identity around JPEGs. It was a social status thing more than a technology thing.

The NFT market recorded $25 billion in total volume in 2021. OpenSea captured over $14 billion of that, more than 60% of the entire market. By January 2022, OpenSea hit a record $5 billion in monthly trading volume. The same platform that had been quietly indexing CryptoKitties a few years before was now processing more volume than most traditional auction houses combined.

When the market ignored builders

At this moment, I was working as an engineer for Mintbase. It was a company that started in early 2018 with the goal of building utility NFTs (tickets, access passes, anything that could be digitized as a token and actually used, not just traded). We were stubborn and true to the utility vision and were largely left out of the PFP hype.

Until we did a collection with Deadmau5 and processed 1 million NFTs in a few weeks in late 2021.

That’s the insight that stung the most. We had been building utility NFTs since 2018, the “right” vision, and were ignored by the market for three years. When we finally got real traction, it wasn’t because the market matured into our thesis. It was because of a celebrity partnership. The market didn’t want utility. It wanted spectacle.

We were right about what NFTs could be, but the market only showed up for what they actually were in practice: cultural speculation.

This connects directly to Peter’s tweet. You can be technically right for years and still feel like you wasted your time because the market never met you where you were.

This was the moment crypto stopped being a tech narrative and became a cultural one. That brought in millions of new people, but most of them were there for the flip, not the tech. And ready to leave the moment things didn’t go their way.

Personally, I think of crypto as a tech, political, and financial movement. There are beautiful pillars to it: individual freedom, accountability, and empowerment built as hard protocols, not soft promises. But that’s easily missed when quick money grabs make it easy to forget the tech and political movement underneath.

In late 2021, Facebook renamed itself to Meta. That was the signal for an entire narrative shift.

I had been running a VR studio for some time, where we shipped three games to the Oculus stores before any VR hype, one with more than 200,000 downloads. Remember my friend who told me about Ethereum four years earlier at Arco do Cego? He was my co-founder in the studio. We had been excited about spatial computing for years. For us it was clear that VR was a platform of the future, but there were still so many User Experience problems to solve that the Meta announcement caught us off guard.

Here’s what hype does to you. I had actual VR experience. I had shipped products with real users. I knew the UX problems firsthand: the motion sickness, the clunky controllers, the isolation, the limited content library. And when a trillion-dollar company announced they were betting everything on the metaverse, even I, someone with more hands-on VR experience than 99% of crypto people, started second-guessing my own expertise. Maybe they had insights we didn’t.

They didn’t. Meta has since quietly walked back most of its metaverse ambitions. The lesson: trust the builder’s instinct over the corporate narrative. If you’ve been in the trenches and something doesn’t feel ready, it probably isn’t, no matter who announces otherwise.

Decentraland, The Sandbox: virtual worlds where you could buy plots of digital land as NFTs. People were paying hundreds of thousands of dollars for virtual real estate that barely anyone visited. Brands like Adidas and Samsung opened virtual stores.

The metaverse thesis was that we would spend most of our lives in virtual worlds and need digital property, fashion, and identity. The reality was that most of these platforms had a few hundred concurrent users at best.

This was probably the peak of the gap between narrative and reality in crypto. The money pouring in was enormous. The actual usage was almost nonexistent.

Around November 10, 2021, Bitcoin hit $69,000 and Ethereum reached $4,867. The total cryptocurrency market cap peaked at $3.03 trillion. Everyone was a genius again. People were quitting their jobs to trade NFTs full time. Crypto Twitter was unbearable.

In hindsight, this was the top. And looking back, we were all possibly a bit affected by months of Covid curfew. There’s something about being locked inside with a screen and a brokerage account that makes people feel invincible.

The failures were human, not technical

2022 was the year crypto ate itself.

In May, Terra/Luna collapsed. It requires a brief explanation of how it worked, because the mechanics matter.

UST was an algorithmic stablecoin designed to always be worth $1, but it wasn’t backed by actual dollars. Instead, it maintained its peg through an arbitrage mechanism with a sister token called LUNA: if UST went above $1, you could mint new UST by burning $1 worth of LUNA (increasing supply, pushing price down). If UST went below $1, you could burn UST to mint $1 worth of LUNA (decreasing supply, pushing price up).

The problem: LUNA’s value partly depended on UST demand. And UST demand was almost entirely driven by one thing: Anchor Protocol, which offered 19.5% APY on UST deposits. That yield was subsidized, not earned. At its peak, Anchor held roughly $14 billion in UST, about 72% of all UST in circulation.

The entire system was circular. UST’s value depended on LUNA. LUNA’s value depended on UST demand. UST demand depended on Anchor’s yield. Anchor’s yield depended on subsidies. When confidence broke, the reflexivity would work in reverse.

In early May 2022, large withdrawals from Anchor triggered a loss of confidence. The mint/burn mechanism kicked in, creating a death spiral: people burned UST, minting massive amounts of LUNA, crashing LUNA’s price, which further destroyed confidence in UST. The Luna Foundation Guard deployed its Bitcoin reserves to defend the peg. It wasn’t enough.

Within days, LUNA had dropped from $80 to under $0.001. Its supply had hyperinflated from 340 million to trillions. UST was at $0.15. The Terra blockchain was halted.

$40-45 billion in combined market cap evaporated in a week. People lost their life savings. Some lost more than that.

This wasn’t just a market failure. It was the logical endpoint of crypto’s obsession with self-referential systems. Real yield comes from external demand. Everything else is redistribution disguised as returns.

The collapse of Terra was like pulling a thread. Every overleveraged entity in crypto started falling.

Three Arrows Capital (3AC), one of the most prominent crypto hedge funds, went bankrupt. They had been borrowing from everyone and putting it all into the same concentrated bets, including hundreds of millions in LUNA.

Celsius, a crypto lending platform that promised retail users high yields on their deposits, froze withdrawals on June 12, 2022. Customer claims totaled $4.7 billion. Voyager filed for bankruptcy. BlockFi followed.

And then November. FTX.

On November 2, 2022, CoinDesk reporter Ian Allison published an article that would unravel the second-largest crypto exchange in the world. He had obtained Alameda Research’s balance sheet, and it revealed that the trading firm’s assets were largely composed of illiquid tokens from its own exchange’s ecosystem, including billions in FTT, a token that FTX itself had created. It was circular, the same self-referential structure that killed Terra/Luna.

On November 6, Binance CEO CZ tweeted: “Liquidating our remaining FTT.” The announcement triggered a bank run. $6 billion in withdrawals hit FTX within 72 hours.

FTX halted withdrawals on November 8. CZ signed a non-binding letter of intent to acquire FTX, then pulled out the next day: “The issues are beyond our ability to help.”

On November 11, FTX and roughly 130 affiliated entities filed for Chapter 11 bankruptcy. John J. Ray III, the lawyer who had overseen Enron’s bankruptcy, was appointed and stated: “Never in my career have I seen such a complete failure of corporate controls and such a complete absence of trustworthy financial information.”

Sam Bankman-Fried had been using customer funds to prop up Alameda Research through a software backdoor. $8 billion in customer deposits were missing. He was arrested in the Bahamas, convicted on all seven counts of fraud, and sentenced to 25 years in prison.

This was not a smart contract bug. This was not a protocol failure. This was plain fraud, the kind of fraud crypto was supposed to make impossible through transparency and decentralization.

But FTX was centralized. People trusted a person, not a protocol.

The entire promise of crypto was trustless systems. And the biggest blowup happened because millions of people sent their money to a centralized entity run by one guy. Self-custody existed. Decentralized exchanges existed. People chose convenience over the principles the technology was designed to enforce.

The protocols I believed in (Uniswap, Ethereum, Aave) kept working through all of it. They never went down. They never froze withdrawals. The code did what it was supposed to do. The failures were human, not technical.

That’s the builder’s frustration. You build the engine, someone else crashes the car, and the headline is “cars are dangerous.”

The wrong user

Towards the end of 2022, there was a new thing in town called ChatGPT. I had been following OpenAI and GPT for some time. But it was only in early 2023 that function calling became available, and to me, that was the real inflection point.

Function calling meant that AI models could interact with the world programmatically. Not just generate text, but trigger actions. Call APIs. Build transactions. That changed everything for a builder like me.

I built very early experiments: ChatGPT constructing blockchain transactions, interacting with smart contracts, minting NFTs, deploying contracts, all from inside the AI. I was watching a language model construct the actual hexadecimal transaction data for an on-chain operation. It worked.

OpenAI launched ChatGPT plugins in March 2023, initially in limited alpha, then broadly to Plus users by May. I published my blockchain interaction experiments through the plugin store.

They got taken down. OpenAI’s review guidelines restricted what they called “financial plugins,” anything that facilitated transactions or financial operations. The reality is that the only “financial” part was paying a few cents in gas to mint an NFT or deploy a contract, the same kind of micro-transaction that any API call involves.

This was a revealing moment. The AI-native companies were not as pro-crypto or open to financial innovation as I’d expected. They saw blockchain = finance = regulated = risk. They allowed the APIs to be used for the same operations, just not through their consumer-facing store. The platform controlled distribution, not technology.

That rejection was important. It pushed me to build independently rather than on someone else’s platform. Sound familiar? It’s the same lesson crypto has been teaching for a decade.

So I built a demo: an early version of a wallet focused on AI actions. Ask your wallet “I want to maximize my yield” and it would analyze available protocols, build a strategy, construct the transaction, and propose you sign and submit it. The AI handled the complexity. The human kept authority. The blockchain handled trust.

This was the moment the two worlds clicked together for me. The gas problem I identified as a broke college student, making users pay for compute, finally had a natural resolution. AI agents don’t mind paying for compute. That’s literally what they do. The UX problem that plagued crypto for seven years wasn’t a problem at all when the user was software.

Taking ownership of this vision and being a hard-working lead engineer led my CEO to promote me to CTO, and to own the full development of what became bitte.ai. The thing I saw at the intersection of AI and crypto became the company’s direction.

2023 was mostly quiet in crypto. Most people declared it dead. Again. The mainstream narrative was that it was all a scam, always had been. The builders who stayed were building, but it felt like 2018 all over again. Except this time the industry had real scars. Real people lost real money to real fraud, not just speculative losses.

The irony is that through all of it, the protocols that were actually decentralized kept working. Uniswap never stopped. Ethereum never went down. Aave kept processing loans. The code did what it was supposed to do.

And while nobody was watching, the infrastructure kept improving dramatically:

Transaction costs collapsed. In March 2024, EIP-4844 (“proto-danksharding”) went live, reducing Layer 2 transaction costs from dollars to fractions of a penny. On networks like Base (Coinbase’s L2), a transaction costs less than $0.01. Solana processes transactions for about $0.00025.

Ethereum moved to Proof of Stake. On September 15, 2022, the Merge happened, the most complex live infrastructure upgrade in crypto history. The Beacon Chain, which had been running in parallel since December 2020, merged with the execution layer. Mining was eliminated entirely. Energy consumption dropped by 99.95%. ETH issuance fell roughly 90%.

Account Abstraction arrived. ERC-4337, deployed to Ethereum mainnet on March 1, 2023, enabled smart contract wallets as first-class accounts. No more seed phrases; wallets could use social recovery, biometrics, or passkeys. “Paymasters” could sponsor gas fees or let users pay in stablecoins instead of ETH. This was the UX fix crypto had needed since 2015.

Developer tooling matured. Foundry replaced Truffle. Viem and wagmi replaced web3.js. Base launched in August 2023 as Coinbase’s Layer 2.

The rails got better while the reputation got worse. My crypto winter was my AI spring, building experiments, understanding models, figuring out what these systems could actually do when connected to real infrastructure.

In January 2024, something unexpected happened. The SEC approved spot Bitcoin ETFs. BlackRock, Fidelity, the biggest names in traditional finance, were now offering Bitcoin as a product. The same institutions that had dismissed crypto for years were selling it to pension funds.

Bitcoin started climbing again. Not because of a new narrative or a new meme. Because Wall Street showed up.

This was a turning point. Crypto was no longer just for the degens and the dreamers. It was an asset class. Whether you liked it or not.

While Bitcoin was going institutional, the other side of crypto went full degenerate.

Solana, a blockchain that most people had written off after FTX (the exchange and Alameda Research were deeply tied to the Solana ecosystem) came back from the dead. SOL had crashed from about $35 before the FTX collapse to roughly $8 by late December 2022. For most of 2023, it traded between $15 and $25. By late December 2023, SOL was above $100. By March 2024, it hit $200.

What drove the recovery? Memecoins.

Dogwifhat, BONK, and eventually a tsunami of tokens with no utility, no roadmap, no pretense. Just vibes and speculation.

The accelerant was pump.fun, which launched in January 2024. It let anyone create a Solana token in seconds for less than $2, no code required. A bonding curve handles initial pricing; when market cap hits roughly $69,000, liquidity automatically migrates to the Raydium exchange. By early 2025, over 7 million tokens had been created on the platform. Revenue exceeded $250 million.

The failure rate is staggering. Only about 1.4% of tokens ever “graduate” to the exchange. Over 98% effectively go to zero. Every 24 hours, roughly 10,000 tokens are launched and nearly 10,000 become defunct.

This was crypto stripped to its most honest and most absurd form. No one was pretending these were revolutionary technology. It was a casino and everyone knew it.

I personally think memecoin markets are as interesting as prediction markets. Memecoins are attention markets, the same way Instagram and TikTok capitalize on your attention, but with memecoins we have direct access to the mechanics at play. Instagram monetizes attention through ads; you are the product. Memecoins monetize attention through direct participation; you are the market. One is hidden extraction, the other is transparent gambling. Neither is virtuous, but at least memecoins are honest about what they are.

Vitalik Buterin wrote a thoughtful piece arguing that memecoins represent an untapped opportunity. While current iterations are harmful, the underlying appeal (fun, accessibility, financialization) could be redirected toward positive-sum outcomes. He proposed charity coins and “Robin Hood games” that combine entertainment with public goods funding.

The problem is that regular internet users don’t see any of this nuance. All they see is dopamine websites and rugpulls. And they’re not entirely wrong. The most visible use cases keep confirming their worst assumptions about crypto.

Maybe stablecoins should have been introduced earlier in this story, because they quietly became the most successful crypto product while everyone was arguing about everything else.

From roughly $20 billion in market cap in early 2020, stablecoins grew to over $320 billion by early 2026. USDT (Tether) leads at about $184 billion, USDC (Circle) at $79 billion.

In 2025, stablecoins settled $33 trillion in transaction volume, surpassing Visa and PayPal combined. B2B stablecoin payments surged 30x, from under $100 million per month in early 2023 to over $3 billion per month in 2025. USDC now captures about 64% of adjusted transaction volume, overtaking USDT for the first time despite having less than half its market cap.

In Latin America, 71% of survey respondents use stablecoins for cross-border payments, replacing traditional remittances that still cost an average of 6.49% per transfer. The IMF published a paper titled “How Stablecoins Can Improve Payments and Global Finance.”

Stripe acquired Bridge, a stablecoin infrastructure company, for $1.1 billion, the largest crypto acquisition in history. BlackRock launched BUIDL, a tokenized Treasury fund, on Ethereum in March 2024. PayPal launched its own stablecoin, PYUSD.

Stablecoins are arguably what crypto was always supposed to build: a way to move money globally, instantly, cheaply. No volatility. No speculation. Just infrastructure.

Crypto seemed to be losing its momentum to AI. AI became the default conversation in tech. Every industry scrambled to figure out what it meant for them. Crypto was no exception.

Projects explored AI agents with wallets, decentralized compute for model training, token-incentivized data collection. Some of it was genuinely interesting. A lot of it was two hype cycles duct-taped together.

But something real was happening underneath.

LLMs improved dramatically. By early 2025, models were capable of long-horizon tasks, not just answering questions but planning, executing multi-step workflows, and interacting with external systems. The function calling I got excited about in 2023 evolved into full agent capabilities.

And agents need to pay for things.

The user arrived

Here’s the concrete example that made this real to me. Cloudflare, one of the largest web infrastructure companies, launched AI Crawl Control. On an average day, Cloudflare customers send over 1 billion HTTP 402 “Payment Required” responses to AI crawlers. Every day. Over a billion times.

402 is an HTTP status code that was written into the original web specification in 1997 with four words: “reserved for future use.” The web anticipated that one day, software would need to pay for things natively. But credit cards won instead. Subscriptions dominated. And 402 sat dormant for 28 years.

Now AI agents need to buy things. And 402 is finally waking up.

In September 2025, Coinbase and Cloudflare shipped x402, a protocol that activates that dormant HTTP code for stablecoin micropayments directly over HTTP. An AI agent hits an endpoint, gets a 402 response with payment details, signs a USDC payment for fractions of a cent, retries with the payment proof, and the response comes back. No account. No subscription. No human in the loop. Sub-second.

Over 100 million transactions in the first months. 10,000% transaction volume growth in a single month. Google integrated it into their agentic payments protocol. Visa, Anthropic, and Cloudflare co-founded the x402 Foundation. $600 million in annualized volume.

And here’s the tell: on February 11, 2026, Stripe launched x402 support. They’re using USDC on Base. The same rails. The same protocol. The same week, Coinbase launched Agentic Wallets, the first wallet infrastructure designed specifically for AI agents.

Stripe isn’t competing with crypto. Stripe is building on crypto.

When the biggest payment company in the world quietly adopts the protocol, the question stops being “crypto or traditional finance.” The question becomes: which layer do you want to build on, the platform-controlled one, or the open one?

Google has AP2, their agentic payments protocol. It requires approval. It’s gated. OpenAI is building delegated payment. Same model, platform-controlled.

x402 is different. Anyone can implement it. No one controls it. It’s the HTTP of payments.

Speaking at AI Hub Lisboa

What I’m building now

I’m building frames.ag, financial infrastructure for AI agents.

Everything in my journey led here. The broke college student who wanted free cloud compute. The builder who spent years shipping NFT infrastructure the market ignored. The engineer who got his ChatGPT plugins banned. The CTO who built an AI wallet at Bitte. The person who watched crypto solve every technical problem while failing every product test.

That’s why I’m building Frames. Not because I think crypto wins as a brand, but because I think open programmable payment rails finally have a user that actually wants them.

This isn’t crypto for vibes. It’s crypto for work.

Peter Steinberger said don’t waste your 20s on crypto. The irony is that crypto scammers immediately proved his point by launching a fake token in his name.

Every crypto cycle was the right idea at the wrong time, built for the wrong user. Digital scarcity, programmable money, permissionless payments, composable finance. These were real technical achievements. But they were pointed at humans who didn’t want to deal with seed phrases and gas fees and 12-word recovery phrases.

The cypherpunks wrote code. They built the cryptographic foundations. Bitcoin proved digital scarcity. Ethereum proved programmable money. DeFi proved composable finance. Stablecoins proved digital dollars. The infrastructure works. It always worked. The protocols never crashed; people did.

In 1999, a trillion and a half dollars went into telecom. Most of those companies died. But they laid fiber under every ocean. That fiber powered YouTube, Netflix, and Zoom. The bubble was the business model for the infrastructure.

Crypto’s version of that story is playing out right now. The tourists left. The builders stayed. The infrastructure is better than it’s ever been: sub-cent transactions, $320 billion in stablecoins, $33 trillion in annual settlement volume. And now there’s a new user, AI agents, that can’t open bank accounts but need to transact billions of times a day.

HTTP 402 sat dormant for 28 years, reserved for future use. The future arrived.

I spent my 20s learning how money moves through code, how trust works without intermediaries, and how to build products when nobody is using them yet. AI agents need all of that.

I didn’t waste my 20s. I just didn’t know who I was building for yet.

This essay was also presented as a talk.