Tony Chen

@naval 💯

The new competition isn’t Humans vs AI. It’s Humans with AI vs everyone else.

stablecoins will save cryptos reputation

an agent making a payment is not impressive anymore an agent being allowed to make a payment is

stablecoins are one of those technologies that look niche right before they look obvious

@circle @jerallaire programmable money is step one programmable commerce is where things get interesting

Programmable money is the starting point. The bigger opportunity is the network around it. @jerallaire on how digital dollars, digital euros, and open infrastructure begin to scale into a much larger financial system.

crypto cards for humans today agent wallets tomorrow

@AriEiberman imo maybe the problem is not how do we remove KYC from cards. the problem is that existing payment rails were designed around persistent human identity, while AI-native commerce is moving toward delegated permissions, bounded execution, and programmable trust.

Non-KYC card issuance sounds exciting and I see the lure of it. But the reason why they can’t exist (today) goes beyond the user data. At least not in the way most people think. I get asked about it regularly at Gnosis Pay. And I understand the appeal, everyone can access a global payment network. But here's what most people miss when they bring it up. There is no such thing as 100% KYC-less card issuance in compliant infrastructure. What people usually call "non-KYC cards" are pre-paid/Business cards. Full stop. And when you look at corporate cards, the KYC just moves up the chain. The business does KYB. Leadership does KYC. Employees get verified internally. So when you are not in the payroll using that card, that’s infringement. The information has to exist somewhere, because the moment fraud hits, someone needs to be accountable. Remove that entirely and the whole thing falls apart. I'm not naive about why people want this. The appeal of private, frictionless spending is amazing. and it's one of the most interesting unsolved problems Web3 has on its plate. But it will not be solved through Visa, Mastercard, or Amex. That's not a criticism, it's just how traditional card networks are built. If non-KYC card spending is ever going to be a legitimate, scalable option, someone needs to build an entirely new card network - one that is globally accepted, lighter on identity requirements, and with fraud prevention so strong that fraud essentially stops before it happens. Because right now, the moment fraud enters the picture, every non-KYC promise shatters. We need a card network that is fraud-resistant first. Then talk about KYC-less cards. That's the real product waiting to be built.

The internet created a world where software could publish information, distribute media, and coordinate communication at planetary scale. But one thing remained fundamentally human: economic agency. Payments, credit, banking, and financial trust were all built around the assumption that a person sits behind every action. Every layer of modern finance, from cards to compliance systems, implicitly assumes human intent. That assumption is starting to break. AI agents are no longer just generating text or retrieving information. They are beginning to execute workflows, negotiate with services, manage resources, purchase compute, coordinate tasks, and eventually transact with other agents autonomously. What emerges from this is not simply "AI-assisted payments," but an entirely new class of economic participant. The existing financial stack is not designed for this world. Human-centric interfaces like OTPs, banking apps, checkout flows, and manual approvals become friction when software itself becomes the actor. The challenge is not just enabling agents to move money. The real challenge is establishing whether an autonomous system should be trusted to act financially in the first place. In the coming decade, financial infrastructure will increasingly revolve around machine-native identity, programmable permissions, verifiable behavior, autonomous treasury management, and agent reputation systems. Stablecoins will likely become the default settlement layer not because they are "crypto," but because they are the first globally accessible form of programmable internet money that software can use natively. The next era of finance will not simply digitize existing institutions. It will introduce entirely new economic actors into the system. And once software becomes capable of holding capital and making decisions, programmable trust becomes more important than programmable money. That is the world @KleePay is being built for.

next-gen payments is underselling what’s actually happening. the bigger shift is that we’re entering a world where software itself becomes an economic actor. AI agents will not operate inside banking apps, card checkouts, or financial workflows designed for humans, they will transact directly across APIs, services, models, and machines. that changes what payment infrastructure needs to look like. stablecoins matter not just because they make payments faster, but because they’re one of the first forms of money that feels truly native to the internet.

Open Sourcing Centaur: Multiplayer, self-hosted, secure agents for Slack. Centaur has been transforming how @paradigm and @tempo invest, build and research. Now you can run it yourself on infrastructure you control. Instructions below.

crypto is no longer one industry — it's at least 4: 1. stablecoins + payments 2. Bitcoin, crypto asset class 3. tokenization + onchain financial services (defi) 4. blockchain infrastructure they are of course inter-related. but increasingly divergent in context which is part of the mixed vibe right now

Google has quietly dropped what researchers are calling "Attention Is All You Need V2." And it signals the end of the Transformer era as we know it. In 2017, the original "Attention Is All You Need" paper changed the world by proving that AI doesn't need recurrence, it just needs to pay attention. But today, even the most advanced models like GPT and Gemini suffer from a massive, structural flaw: Catastrophic Forgetting. The moment an AI learns something new, it starts losing what it learned before. It’s why AI "hallucinates" or loses the thread in long conversations. This paper, titled "Nested Learning: The Illusion of Deep Learning Architectures," completely replaces the way AI stores information. The researchers have introduced a paradigm shift called Nested Learning (NL). Here is why this is "V2": For the last decade, we treated AI models as one giant, flat mathematical function. NL proves that a model is actually a set of thousands of smaller, "nested" optimization problems running in parallel. Instead of one giant "memory," each layer has its own internal "context flow." This allows the model to learn new tasks at test-time without overwriting its core intelligence. It moves us past the static Transformer. The new architecture (HOPE) demonstrated 100% stability in long-context memory and "post-training adaptation" that was previously impossible. The technical takeaway is brutal for the competition: Existing deep learning works by compressing information until it breaks. Nested Learning works by organizing information so it can grow forever. We’ve spent 7 years trying to make Transformers bigger. Google figured out how to make them "Nested." The Transformer replaced the RNN in 2017. Nested Learning is here to replace the Transformer in 2026.

There’s never been a better time to become a founder. Just do it.

Agent payments will generate real fee flow at scale, just like card payments have over the past 50 yrs. @KleePay is positioning itself in that flow from day 1, much like @Stripe did in the card ecosystem back in 2010.

Fireside chat at Sequoia Ascent 2026 from a ~week ago. Some highlights: The first theme I tried to push on is that LLMs are about a lot more than just speeding up what existed before (e.g. coding). Three examples of new horizons: 1. menugen: an app that can be fully engulfed by LLMs, with no classical code needed: input an image, output an image and an LLM can natively do the thing. 2. install .md skills instead of install .sh scripts. Why create a complex Software 1.0 bash script for e.g. installing a piece of software if you can write the installation out in words and say "just show this to your LLM". The LLM is an advanced interpreter of English and can intelligently target installation to your setup, debug everything inline, etc. 3. LLM knowledge bases as an example of something that was *impossible* with classical code because it's computation over unstructured data (knowledge) from arbitrary sources and in arbitrary formats, including simply text articles etc. I pushed on these because in every new paradigm change, the obvious things are always in the realm of speeding up or somehow improving what existed, but here we have examples of functionality that either suddenly perhaps shouldn't even exist (1,2), or was fundamentally not possible before (3). The second (ongoing) theme is trying to explain the pattern of jaggedness in LLMs. How it can be true that a single artifact will simultaneously 1) coherently refactor a 100,000-line code base *and* 2) tell you to walk to the car wash to wash your car. I previously wrote about the source of this as having to do with verifiability of a domain, here I expand on this as having to also do with economics because revenue/TAM dictates what the frontier labs choose to package into training data distributions during RL. You're either in the data distribution (on the rails of the RL circuits) and flying or you're off-roading in the jungle with a machete, in relative terms. Still not 100% satisfied with this, but it's an ongoing struggle to build an accurate model of LLM capabilities if you wish to practically take advantage of their power while avoiding their pitfalls, which brings me to... Last theme is the agent-native economy. The decomposition of products and services into sensors, actuators and logic (split up across all of 1.0/2.0/3.0 computing paradigms), how we can make information maximally legible to LLMs, some words on the quickly emerging agentic engineering and its skill set, related hiring practices, etc., possibly even hints/dreams of fully neural computing handling the vast majority of computation with some help from (classical) CPU coprocessors.

the future of payments belongs to stablecoin card.

turned 30, left a top-tier VC job to start something not because i wanted to “be a founder” but because after seeing enough companies up close, i stopped guessing i know what matters and what doesn’t it’s less a leap, more acting on something that already felt obvious

AI is making intelligence cheaper and taste more expensive.

Every agent will need its own computer. And with new Hosted agents in Foundry, every agent gets its own dedicated enterprise-grade sandbox, with durable state, built-in identity and governance, and support for any harness or framework. Read more: devblogs.microsoft.com/foundry/introd…