CTO @ZKcandyHQ Founder @ceo__ai author of ERC-7662 draft standard for AI Agent NFTs

Joined November 2011
64 Photos and videos
Pinned Tweet
18 Feb 2025
Big things coming soon on @ZKcandyHQ - can't wait to show what we've been up to.
3
1
13
5,159
Serious question for founders running on 4 hours of sleep: What percentage of your work last week was stuff only YOU could do? Not "stuff you did." Stuff that specifically required your brain, your relationships, your judgment. The rest? That's agent work. And you're doing it by hand.
1
1
65
24 Nov 2025
If you looking for a job in software development easiest way to level up your skills and build your network that can lead to a job is hackathons (and you can win prizes). Some orgs to look at: @ethglobal @encodeclub @devpost
2
807
19 Nov 2025
Feeling frisky. This Agent is going to be a lot of fun to work with.
3
1
3
2,574
Greg Marlin retweeted
We have identified the likely cause of the incident and are working on a fix. We will provide another update as we get closer to deploying the fix. githubstatus.com/incidents/5…
1
16
4,197
16 Nov 2025
Do you remember when you joined X? I do! #MyXAnniversary wow 14 years
1
2
366
Greg Marlin retweeted
It’s nearly 2026, and billions of people still have to wait for the US to wake up before they can trade the best financial markets. Trading hours and markets that close are outdated. Tokenized assets will be better for everyone, with instant settlement and 24/7 availability.
756
807
6,796
436,938
Greg Marlin retweeted
30 Oct 2025
The long wait is almost over. New website coming soon. If you want to secure your spot to first wave codes make sure to sign up at CEO.ai
2
3
606
30 Oct 2025
For when you need to finish a long session strong, I highly recommend Mo'Horizons youtube.com/watch?v=fhnTkj0G…

1
562
30 Oct 2025
Built my first Electron app that I use every day. It's a simple time tracker with categories and reporting (runs locally data is all local - no callouts). If you think I should open source it drop a comment, otherwise I'll keep it closed.
195
Greg Marlin retweeted
🚨 JUST IN: JPMorgan CEO Jamie Dimon admits he was WRONG about Bitcoin and crypto "It's real." "It will be used by all of us." 😲
574
1,365
9,877
918,155
30 Oct 2025
Tonight's session is going with the real full 8-hour trance mix from Delta Notch. youtube.com/watch?v=_2ZIvK44…

203
Greg Marlin retweeted
28 Oct 2025
ai agents will become one of the most valuable assets to own on the internet
10
2
28
1,014
28 Oct 2025
Starting the home stretch of this session with Tiga Sunglasses at Night seems appropriate. youtube.com/watch?v=ML7wMEjU…

245
28 Oct 2025
Tonight's hacking session is starting early and going to go all night, which means starting mellow with some Carbon Based Lifeforms. youtube.com/watch?v=XWNsWZ5V…
1
246
Greg Marlin retweeted
25 Oct 2025
The x402 Credit Card Summary: Traditional credit cards don’t work for agentic commerce. They aren’t programmable. They don’t handle micro-transactions cost-effectively. They don’t integrate easily with agent payment rails like x402 standard and MCP servers. And most consumers will never prefund a crypto wallet so that an AI agent can make micro-transactions. If agents are going to shop for us and drive the future of e-commerce, we need a purpose-built “credit card” for agents. But this isn’t a legacy card from legacy players. It needs to be fully on-chain, API-based, with a shared pool of liquidity that agents can pull from to make payments, and the human owner pays back on an interval (i.e. BNPL-style repayment). Let’s dig in. AI Agents Need a Purpose-Built Credit Card. AI agents will shop for us. They’ll order groceries, book travel, schedule dinner at exactly 6:15 pm. But to do so, they’ll also need to purchase access to data, in micro, sub-penny transactions, in order to hunt for the right items. For example: “check live inventory, query size availability, get ratings” might cost a few hundredths of a cent per query. The emerging MCP x402 infrastructure is built for this: an MCP server exposes a tool, the agent hits it, the server issues HTTP 402 payment requirement, the agent pays via stablecoin (e.g., USDC) on-chain, then gets the data. However, this existing workflow still demands that the agent (or its wallet) be pre-funded. That means the human must deposit funds into a wallet, or the agent must hold a balance. Most consumers will never do this. That fundamental friction limits agentic commerce, and presents an exciting opportunity to build a new kind of consumer/agent credit product. By introducing a “x402 credit card for agents”, we remove the need for prefunding, enabling both (1) the streaming micro-transactions where the agent buys data, and (2) the everyday spend where the agent purchases goods/services (e.g., a TV), both within the same payment method. The “x402 card” provides agents with a credit line drawn from a shared on-chain pool of liquidity. The agent pings the pool, pulls funds just-in-time, settlement is done in stablecoins, and the human pays back the loan ala BNPL-style repayment (e.g., Pay in 4). The x402 Credit Card To begin, this on-chain “credit card” isn’t an actual card. Its a wallet-to-wallet transaction (agent wallet/pool wallet ➞ merchant wallet). But the key difference: the agent doesn’t pull from a prefunded account. Instead, it triggers the shared pool of liquidity (smart-contract managed), which acts like the issuer. The flow supports two distinct payment types: - Pico/Micro-Transactions: Millions of tiny streaming payments (e.g., $0.001 per API call, data query) — similar to the use-cases targeted by x402. - Everyday Spend: Traditional value transfers (e.g., buying a TV, paying for a concert) settled in stablecoin via the smart-contract infrastructure. Here’s how it works (high-level): Lenders deposit USDC into a pool. The pool uses on-chain yield-optimizers and earns 6-10%, majority of which is returned to lenders to incentivize lending into the pool. This creates a virtually free cost of capital. So, now we have a pool of liquidity. To make a payment, the agent (via SDK) triggers the smart contract credit pull, a “just-in-time” credit draw from that pool. The merchant is settled to instantly in stablecoins (or LP tokens to keep funds in the pool generating yield up until the moment of off-ramping). No traditional card rails. No traditional fee structures. And the human owner repays with BNPL style flexibility (e.g., in four installments), replenishing the pool. This opens the door for both micro-transactions and everyday spend on the same payment product, exactly what agentic commerce requires. How This Integrates with AI Agent Infrastructure Each agent gets an “account ID” linked to a KYC’d human owner account, who sets limits and repayment rules. The agent transacts independently within that limit, without requiring a prefunded wallet. When an MCP server routes a tool request (such as “fetch data”), and that endpoint is x402-protected (HTTP 402), an SDK triggers the flow: - detect 402 - generate payment instruction - draw credit from pool - settle to merchants in stablecoin/LP tokens. For larger e-commerce spend (non-tool-call), the same API triggers the payment logic and settlement. The result: an agent can pay $0.001 to query a data API, then pay $1,000 to buy a laptop, using the same payment infrastructure. No prefunded wallet required. MCP servers act as middleware between AI agents and external services (APIs, data, commerce endpoints). Right now, MCP mostly routes queries/responses; when money is needed, the lack of a seamless payment method is a bottleneck. The x402 protocol is designed to be that payment method, machine-native, stablecoin-based, minimal friction. An on-chain credit card will act as a “payment method” visible to MCP servers and agents: when an x402-protected endpoint returns HTTP 402, our SDK executes the credit-pool draw and settlement to the merchant. Fundamentally we combine: MCP tool invocation x402 payment handshake on-chain credit draw stablecoin settlement human repayment. Credit Card PCI Compliance Nightmares AI agents holding credit card PANs (card numbers) is a PCI compliance and regulatory nightmare. Tokenization of cards is being proposed, but it remains constrained (domain use, lifecycle issues). With our system, agents never hold card numbers, they hold wallet keys or agent IDs. The MCP server and smart contract layer handles authorization. Each agent gets a virtual credit line tracked on-chain, with spend limits, risk models and repayment schedule. Humans pay back later; agents transact independently. Visa Intelligence Commerce Here is where some might say, but Phil, Visa Intelligent Commerce meaningfully reduces friction and risk for AI agents. Sure, it provides agent-bound tokenization, controls and developer tooling so agents can transact more safely and with better user controls. All true. But it does not solve the core economic and technical gaps that agentic commerce requires at scale: ultra-low-cost micropayments, native programmability/composability for payments, and ledger-anchored instant settlement. Visa’s solution remains an evolutionary extension of existing card rails (and therefore inherits their fee and settlement economics), whereas an on-chain credit primitive, a pooled, smart-contract based stablecoin credit line that agents can draw from and that settles instantly on-chain, offers a fundamentally different set of economic tradeoffs that unlocks micro-transactions and dramatically lower merchant costs. Why Agentic Commerce Needs to Be On-Chain Legacy rails (Visa/Mastercard) are not engineered for what agents need. They can’t auth micro-payments: they batch, round, have high fixed costs, minimum fees. They aren’t programmable (you can’t compose logic/triggers into card payments easily). And they don’t integrate with x402. An x402, On-Chain “Credit Card” Solves This With: Micro-transaction readiness (tiny payments baked in). Pool-based liquidity funded by lenders (yield-earning until used). Smart contract based just-in-time credit pulls (instant authorization). Programmable APIs compatible with MCP/x402 flows (agents can script payments). Secure, composable fund flows (agent → pool draw → merchant settlement). AI agents will be the next frontier of e-commerce, hunting for items, executing flows, interacting autonomously on behalf of humans. To enable that at scale, they need a payment method built for their paradigm: the on-chain x402 credit card. It enables micro-transactions, seamless authorizations, programmable flows, instant settlement, and doesn’t require prefunding a wallet (big consumer hurdle). If agentic commerce is the future of how we shop online, an on-chain credit card will be the preferred payment method, and is a massive opportunity. And if you read this far, I appreciate you. Would love thoughts. Feel free to reach out. DMs open.
79
42
479
82,935
Greg Marlin retweeted
In the last week x402 has had: 1.12 million transactions $1.24 million in volume 69k buys of 774 services Across >7 facilitators, >3 chains Just getting started.
19
14
153
11,694