Subscribe
The Crypto × AI Agent Stack
Three Axioms for the Next Decade:
AI agents fully displace workforce across major verticals.
AI labor is paid per-task, per-use, shattering the traditional salary model.
Agents settle payments in a global, permissionless currency: stablecoins.
The rails already exist. The opportunity to back companies building at the edge of this transition is enormous.
Agentic Commerce: The Missing Link
Today, agents can browse, reason, code, and even deploy apps autonomously. But the moment they need to pay for something? Dead stop. Agents can’t swipe a Visa. They can’t pass KYC. They can’t sign a contract.
Payments are the missing link between AI capability and AI autonomy.
The infrastructure is being laid at breakneck speed. Circle, Stripe, Visa and Coinbase are all making massive bets despite nascent transaction volumes. Bloomberg recently reported that stablecoin firms are betting big on AI agent payments that, in the words of the headline, “barely exist.”
“We’ll probably overestimate how fast agentic commerce takes off in the next year, but we’re largely underestimating what it can become in five.”
— Artemis Analytics
CZ put it even more directly: AI agents will make one million times more payments than humans. We’re investing in the five-year view.
How Humans Pay vs. How Agents Pay
The fundamental mismatch between legacy payment rails and agent requirements makes the case for crypto-native infrastructure:
“Agents won’t pay like tourists. They’ll pay like locals — through relationships, credit, and repeat business.”
— Sam Broner, a16z crypto
The Protocols
Three giants are building to own the agent payments stack. This is the TCP/IP moment for agent commerce:
Coinbase: x402 (Coinbase), Revives the HTTP 402 status code to enable programmatic payments over HTTP. Launched May 2025, processed 75M transactions. Daily transactions climbed from fewer than 5,000 to a peak of 3.9 million. Coinbase also built the first agent wallet (AWAL), non-custodial wallets in TEEs with CLI setup in under 2 minutes. Open source.
Stripe: ACP (Stripe × OpenAI), Agentic Commerce Protocol, Apache 2.0 licensed. Integrates with Stripe’s $1.9T annual volume. Stripe is also building Tempo, a stablecoin-focused L1 blockchain valued at ~$5B, with Anthropic and OpenAI as initial design partners.
Google: AP2 (Google), Agent Payments Protocol 2.0 with 60+ partners including Coinbase, Circle, MetaMask, and Questflow. Uses x402 under the hood for on-chain settlement.
Stripe’s Five Levels of Agentic Commerce
On their 2025 annual letter, Stripe defined a useful framework for understanding the progression of agent autonomy in commerce:
Level 1: AI fills out forms for the user.
Level 2: AI finds and compares options.
Level 3: AI remembers preferences and context.
Level 4: Full delegation, “buy the best one.”
Level 5: Anticipatory, the agent buys before you even ask.
Cosmo Jiang at Pantera Capital offers a critical insight: Levels 0–4 don’t need new financial rails, a human identity still underlies the transaction. The blockchain layer becomes critical at Level 5 and beyond, where agents transact with each other autonomously, with no human in the loop. That’s the unlock.
The Identity Layer: ERC-8004
Payments require identity. How do you trust a bot you’ve never met?
Identity Registry: Every agent gets a portable, censorship-resistant identifier based on ERC-721 with URI resolution to an agent’s registration file.
Reputation Registry: A standard interface for posting and fetching onchain feedback signals. Scoring and aggregation occur both on-chain (for composability) and off-chain (for sophisticated algorithms).
Validation Registry: Standardizes how agent actions are independently verified onchain, proof that agents actually did what they claim to have done. Uses mechanisms like zkML and TEEs.
The standard was co-authored by MetaMask, the Ethereum Foundation, Google, and Coinbase. It extends Google’s Agent-to-Agent (A2A) protocol with Ethereum’s strengths: trustless verification, cryptographic proofs, and permanent on-chain records.
“x402 and ERC-8004 are establishing crypto as the optimal infrastructure for agentic finance.”
— Chris Davis & Armita, Messari “Decentralized AI” Report
Chris has also updated his views recently mentioning: “all B2C e-commerce will be driven by card for the foreseeable future. You can delegate your card to an agent with a spending limit, which is better for you as the consumer since you get cashback/points/rewards. The real value prop of x402/stablecoins is micro transactions, useful if your agent is hitting some API that you don’t have a key for (data, inference, compute).”
The Startup Explosion
The wallet and payments infrastructure layer is exploding with well-funded startups, many led by former Stripe, Shopify, and Google engineers who understand payments at the deepest level:
ATXP (@atxp_ai) — $19.2M seed. Founded by Louis Amira (Stripe’s former Head of Crypto & AI Partnerships) and David Noel-Romas (Stripe’s former Head of Crypto Engineering). Web-wide protocol for the agent commerce lifecycle, described as “the HTTP for agentic payments.” Backed by Stripe, Coinbase Ventures, Solana Ventures, Samsung Next.
Sponge (@spongeoneth) — YC W26. Three ex-Stripe founders (Jae Choi, Rishab Luthra, Eric Zhang) who built Stripe’s stablecoin financial accounts. Agent wallets + business gateway.
Sapiom — $15.75M seed led by Accel, with Anthropic and Coinbase Ventures. Founded by Ilan Zerbib, ex-Shopify engineering director who scaled Shop Pay to $100B+ GMV. Issues payment and identity “envelopes” for each agent.
KiteAI (@GoKiteAI) — $33M total raised. $18M Series A led by PayPal Ventures + General Catalyst. Founded by Chi Zhang (PhD, UC Berkeley). Any PayPal or Shopify merchant can opt-in to accept agent payments via the Kite Agent App Store.
Micropayments: The Paradigm Shift
The subscription model is dying for APIs. Why pay $200/month for an API when your agent only needs $0.003 worth of data? Per-request micropayments unlock a world where agents hit thousands of endpoints, paying fractions of a cent each.
Merit Systems (@merit_systems) — a16z-backed ($10M seed, co-led with Blockchain Capital). Founded by Sam Ragsdale (ex-a16z, ex-Google) and Ryan Sproule (ex-AWS). Turning every API into pay-per-use settled onchain. Building the attribution protocol for open-source.
Orthogonal (@orthogonal_sh) — YC W26. Developer-first x402 skill marketplace. One-click install hundreds of paid API skills directly into Claude Code or any agent harness. Partners with merchants to wrap their APIs in x402 endpoints, then handles distribution to developers. The “app store” for agent-purchasable services.
Breeze (@Breez_Tech) — Allows agents to deploy idle stablecoins into yield positions while waiting to spend. Yield management for the agent economy.
Autonomous Companies: Agents as CEOs
Perhaps the most striking demonstration of the agentic economy isn’t payments infrastructure, it’s the emergence of fully autonomous companies. AI agents are not just executing tasks; they’re running businesses, managing treasuries, raising capital, and making strategic decisions without human intervention.
Crypto-Native Autonomous Agents
$ClawdBot / OpenClaw (@openclaw) — Created by Austin Griffith (Ethereum Foundation), ClawdBot is the original autonomous onchain builder. Given a wallet and full autonomy, it writes its own code, deploys its own dApps on Base, and manages its own treasury. Peaked at $34M market cap. The agent vested itself tokens to ensure long-term economic alignment, a remarkable emergent behavior.
$Felix (@FelixCraftAI) — Built on OpenClaw by Nat Eliason, Felix operates as autonomous CEO of The Masinov Company. Felix generates revenue through selling AI setup guides, operating Claw Mart, and custom services. Generated $62,000+ in total revenue within its first two weeks of operation. Burns its own tokens. Maintains a public transparency dashboard tracking every dollar.
$KellyClaude (@ClaudeKelly) — An AI persona tokenized on Base, maintained through the Clawdbot automation layer. Listed on BitMart with ~$6.25M market cap and active trading on Uniswap V4.
$BANKR (@bankrbot) — Autonomously launches, buys, sells, swaps coins, and places limit orders. 117,000+ active wallets. 1M+ messages processed monthly. ~$51M market cap.
$Clanker (
Web2 Competitors: Autonomous AI Businesses
The autonomous company trend isn’t limited to crypto. Web2 startups are proving the model works at scale with traditional payment rails:
Polsia (@bencera_) — Founded by Ben Cera, Polsia is an autonomous AI that plans, codes, and markets companies 24/7. Hit $1.25M ARR in 3 months with 1,000+ companies running autonomously. The AI itself raised capital — conducting 77 investor meetings in 10 days. Ben gave Polsia 10% equity in the company.
Nanocorp (@nanocorp_pro) — Launch autonomous AI companies that hire agents, run tasks on schedules, and report results. Reached $193K ARR in 3 days. Dashboard monitoring for entire AI-operated business empires.
Junior (@hirejuniorso) — First AI employee. Shows as a coworker with their own email, profile, and integrated with most workforce tools to delegate tasks.
Paperclip — Install and orchestrate an agent only company. Manage workflows, roles, objectives and tasks from a unified dashboard.
The crypto-native versions have a structural advantage: autonomous agents need permissionless wallets, onchain identity, and programmable money to transact independently at scale. The web2 versions still rely on human-credentialed payment rails, which will constrain them as agent-to-agent commerce grows.
DeFAI: Agents Making Money Today
The intersection of AI agents and DeFi has rapidly transitioned from proof-of-concept to production. Messari segments the stack into Alpha Agents, Execution Agents, AI-Enabled Abstraction, and Portfolio Optimization. While many early projects have shut down, those that remained heads-down are now shipping real products.
Giza (@gizatechxyz) — The leading DeFAI protocol. Arma agent has autonomously managed $32M in assets across 102K transactions with $3.2B in total transaction volume and 57,000+ individual agents. Expanded from stablecoin yield optimization to complex Pendle strategies. Capital from top crypto firms like Re7.
NumerAI (@numerai) — Crowdsourced quantitative hedge fund. Grew from $60M to $440M AUM. Delivered 25.5% net return with 12 straight positive months. JPMorgan committed $500M — the firm’s largest institutional backer. 10,000+ data scientists compete to predict stock returns.
Nof1 (@the_nof1) — Alpha Arena: a live benchmark where LLMs compete trading real money. Season 1 winner: Qwen 3 Max. Users create agents visually or via natural language.
CrunchDAO (@crunchdao) — 47.5% growth in participants (7,800 to 10,400). Global network of ML engineers competing for enterprise clients. Trusted by the Eric and Wendy Schmidt Center and Massachusetts General Hospital for metabolic disease ML competitions.
AI Co-Pilots and Intelligence Layers
True North (@get_truenorth) — AI co-pilot synthesizing onchain and offchain data to help users navigate markets. Data flywheel compounds defensibility over time.
Surf (@Surf_Copilot) — AI co-pilot for crypto market intelligence. Combines real-time onchain data with natural language queries for actionable insights.
NickAI (@getnickai) — AI platform for building drag and drop portfolios in a compliant and secure way. Designed for institutional traders. Over 30 traders of Galaxy Digital are using it to create trading workflows for equities and tokens.
Warden (@wardenprotocol) — Agent aggregation platform competing to attract the best agent developers through distribution leverage.
Decentralized Training & Models
Beneath every agent is a model. Beneath every model is training infrastructure. The centralized AI labs — OpenAI, Anthropic, Google — spend billions on proprietary training. But a parallel decentralized stack is emerging, powered by crypto incentives to crowdsource compute, data, and training at global scale.
Prime Intellect (@PrimeIntellect) — $70M+ total raised (Founders Fund, CoinFund, Distributed Global). GPU marketplace aggregating global compute for collaborative training. Just released INTELLECT-3, a 106B parameter Mixture of Experts model that achieves state-of-the-art on math, code, and reasoning benchmarks. The protocol distributes ownership of trained models across contributors.
Nous Research (@NousResearch) — $55M raised ($50M Series A led by Paradigm, $1B+ valuation). Open-source model powerhouse with 50M+ downloads of the Hermes series. Operates Bittensor subnets for distributed, unbiased model training. The Cortex subnet generates fresh synthetic data continuously for evaluation.
Gensyn (@gensynai) — $43M Series A led by a16z crypto. Decentralized ML compute network that standardizes execution, verification, and coordination of ML workloads across any device — from data centers to laptops. Cryptographic verification ensures accuracy. Testnet launched with rewards for node operators.
Pluralis (@PluralisHQ) — $7.6M seed co-led by Union Square Ventures and CoinFund. Novel “Protocol Learning” approach that shards model training across multiple nodes, preventing extraction of individual weights. Backed by angel investors including Balaji Srinivasan and Clem Delangue (HuggingFace co-founder).
Bittensor: The King of Darwinian Platforms
Bittensor ($2.1B market cap, 40+ subnets, $20M+ annualized revenue) is the backbone of the decentralized intelligence economy. Each subnet addresses a specific AI task through competitive market dynamics: owners define incentives, miners compete to deliver the best solutions, and validators rank performance. Three subnets stand out:
Ridges (SN62) — Decentralized software engineering agents. Achieved 80% on SWE-bench in just 45 days from zero — beating every open-source competitor and rivaling centralized labs. Winner-takes-all dynamic: the top agent receives 100% of emissions (~$15,000/day). New leaders must improve the previous score by at least 1.5%. Draws top AI and coding talent including researchers from OpenAI and DeepMind.
Templar (SN3) — Large-scale distributed model training. Successfully trained a 1.2B parameter model end-to-end with ~200 GPUs across the network. Covenant-72B is a massive accomplishment for them and for deAI more broadly.
Chutes (SN64) (@chutes_ai) — Decentralized serverless inference platform. 250x growth since January 2025 — now serving 3T tokens/month. 85% cheaper than AWS for comparable inference tasks. Supports LLMs, image, and audio models.
“While leading AI companies offered hundreds of millions in salaries to poach researchers from rival labs, Darwinian networks produced genuinely competitive results through permissionless, meritocratic structures.”
— Messari, “Decentralized AI” Report
Training Data: The Picks and Shovels
Beneath every model is data. The Messari Decentralized AI report highlights a stunning stat: Surge AI reportedly generated $1.4 billion in revenue with a team of just 121 people and no venture funding. Scale AI sold a 49% stake to Meta for $14.8 billion, generating approximately $870 million in annual revenue.
Crypto’s edge? Token incentives to crowdsource data collection at massive scale:
Grass (@getgrass_io) — $0 to $12.8M projected Q4 revenue in one year. Users share unused bandwidth passively to scrape differentiated web data. Repeat business from major AI labs.
Kled (@useKled) — Decentralized data marketplace on Solana. Users sell raw data (photos, videos, documents) to AI companies. $100M valuation. Proof-of-Data mechanism with cryptographic hashes for authenticity.
EarnOS (@earnos_io) — Platform that collects user’s information in campaigns to build a user’s complete profile (uber rides, doordash spend, etc)
Shaga (@playonshaga) — DePIN network turning idle gaming PCs into a distributed cloud gaming platform. Produces Gameplay-Action Pair (GAP) data for training world models. ~$26M estimated data valuation. Context: OpenAI offered $500M to acquire Medal for similar gaming data.
The Smart Contract Security Dilemma
As the Messari report warns, AI is a double-edged sword for smart contracts. While AI-assisted coding democratizes development, it also gives hackers powerful tools for exploitation.
The SCONE-bench study from Anthropic evaluated agents’ ability to exploit 405 real-world smart contract vulnerabilities. The results were sobering: ten models collectively produced turnkey exploits for 51% of the problems, with $550 million in simulated funds stolen.
Octane (@octane_security) — AI-powered security scanning integrated into the developer workflow. Continuously scans contracts during development and proposes automated fixes before code reaches mainnet.
TestMachine — AI-native adversarial testing using reinforcement learning and guided fuzzing. Behaves like an autonomous attacker, searching for novel exploit paths that traditional tools miss.
Forta (@FortaNetwork) — Runtime security: monitoring onchain activity in real-time and blocking high-risk transactions before execution. Adaptive firewall for deployed protocols.
We expect capital allocation toward AI-native defense systems to increase sharply as teams realize static audits alone are insufficient against dynamic AI adversaries.
The Bear Case (And Why We Disagree)
We take the skeptics seriously. The numbers today are underwhelming:
Bloomberg titled their March 2026 piece: “Stablecoin Firms Bet Big on AI Agent Payments That Barely Exist.” x402 processes ~$28K/day in real volume against a ~$7B ecosystem valuation.
Artemis found that 86% of Solana’s all-time x402 activity was “gamed” rather than organic.
But here’s the thing: every major platform play looks like a ghost town before it doesn’t. AWS had negligible revenue for years. Stripe processed $0 before it processed $1.9 trillion. Shopify was a snowboard shop before it was a $100B platform.
The infrastructure is being laid by the most credible teams in fintech. The models are getting smarter every quarter. The agents are coming.
The Micropayment Thesis
Rish from Sponge highlighted a use case that goes beyond API payments: content monetization. As AI agents scrape the web, ad-supported publications like the New York Times see their revenue models collapse. The solution may be micropayments, <$0.001 per page view. If it cost a dollar to tweet, he argued, all that AI slop would die overnight.
Our playbook:
Back the payments primitives: agent wallets, payment protocols, micropayment infrastructure.
Back the identity layer: ERC-8004 ecosystem, reputation systems, validation registries.
Back the picks-and-shovels: training data networks, decentralized training labs, Darwinian intelligence platforms.
Back the autonomous companies: the agents that run businesses, manage treasuries, and create economic value independently.
If you’re building at the intersection of crypto and AI agents, our DMs are open.
Subscribe
Sources, Further Reading and Good Follows:
“Decentralized AI” by Chris Davis & Armita — Messari (the definitive deAI report)
“The Financial Rails of Agentic Commerce” by Cosmo Jiang & S.P. Lehman — Pantera Capital
“Agents Will Pay Like Locals, Not Tourists” by Sam Broner — a16z crypto
“The CoinAlg Bind” by Jay Yu, James Austgen et al. — IC3 / Cornell Tech / Pantera
“Stablecoin Firms Bet Big on AI Agent Payments That Barely Exist” — Bloomberg, March 2026
“There’s a strong case for AI and crypto” — Fortune Crypto, March 2026
SCONE-bench: Smart Contract Security Benchmark — Anthropic (GitHub)
Subscribe
The Crypto × AI Agent Stack
Three Axioms for the Next Decade:
AI agents fully displace workforce across major verticals.
AI labor is paid per-task, per-use, shattering the traditional salary model.
Agents settle payments in a global, permissionless currency: stablecoins.
The rails already exist. The opportunity to back companies building at the edge of this transition is enormous.
Agentic Commerce: The Missing Link
Today, agents can browse, reason, code, and even deploy apps autonomously. But the moment they need to pay for something? Dead stop. Agents can’t swipe a Visa. They can’t pass KYC. They can’t sign a contract.
Payments are the missing link between AI capability and AI autonomy.
The infrastructure is being laid at breakneck speed. Circle, Stripe, Visa and Coinbase are all making massive bets despite nascent transaction volumes. Bloomberg recently reported that stablecoin firms are betting big on AI agent payments that, in the words of the headline, “barely exist.”
“We’ll probably overestimate how fast agentic commerce takes off in the next year, but we’re largely underestimating what it can become in five.”
— Artemis Analytics
CZ put it even more directly: AI agents will make one million times more payments than humans. We’re investing in the five-year view.
How Humans Pay vs. How Agents Pay
The fundamental mismatch between legacy payment rails and agent requirements makes the case for crypto-native infrastructure:
“Agents won’t pay like tourists. They’ll pay like locals — through relationships, credit, and repeat business.”
— Sam Broner, a16z crypto
The Protocols
Three giants are building to own the agent payments stack. This is the TCP/IP moment for agent commerce:
Coinbase: x402 (Coinbase), Revives the HTTP 402 status code to enable programmatic payments over HTTP. Launched May 2025, processed 75M transactions. Daily transactions climbed from fewer than 5,000 to a peak of 3.9 million. Coinbase also built the first agent wallet (AWAL), non-custodial wallets in TEEs with CLI setup in under 2 minutes. Open source.
Stripe: ACP (Stripe × OpenAI), Agentic Commerce Protocol, Apache 2.0 licensed. Integrates with Stripe’s $1.9T annual volume. Stripe is also building Tempo, a stablecoin-focused L1 blockchain valued at ~$5B, with Anthropic and OpenAI as initial design partners.
Google: AP2 (Google), Agent Payments Protocol 2.0 with 60+ partners including Coinbase, Circle, MetaMask, and Questflow. Uses x402 under the hood for on-chain settlement.
Stripe’s Five Levels of Agentic Commerce
On their 2025 annual letter, Stripe defined a useful framework for understanding the progression of agent autonomy in commerce:
Level 1: AI fills out forms for the user.
Level 2: AI finds and compares options.
Level 3: AI remembers preferences and context.
Level 4: Full delegation, “buy the best one.”
Level 5: Anticipatory, the agent buys before you even ask.
Cosmo Jiang at Pantera Capital offers a critical insight: Levels 0–4 don’t need new financial rails, a human identity still underlies the transaction. The blockchain layer becomes critical at Level 5 and beyond, where agents transact with each other autonomously, with no human in the loop. That’s the unlock.
The Identity Layer: ERC-8004
Payments require identity. How do you trust a bot you’ve never met?
Identity Registry: Every agent gets a portable, censorship-resistant identifier based on ERC-721 with URI resolution to an agent’s registration file.
Reputation Registry: A standard interface for posting and fetching onchain feedback signals. Scoring and aggregation occur both on-chain (for composability) and off-chain (for sophisticated algorithms).
Validation Registry: Standardizes how agent actions are independently verified onchain, proof that agents actually did what they claim to have done. Uses mechanisms like zkML and TEEs.
The standard was co-authored by MetaMask, the Ethereum Foundation, Google, and Coinbase. It extends Google’s Agent-to-Agent (A2A) protocol with Ethereum’s strengths: trustless verification, cryptographic proofs, and permanent on-chain records.
“x402 and ERC-8004 are establishing crypto as the optimal infrastructure for agentic finance.”
— Chris Davis & Armita, Messari “Decentralized AI” Report
Chris has also updated his views recently mentioning: “all B2C e-commerce will be driven by card for the foreseeable future. You can delegate your card to an agent with a spending limit, which is better for you as the consumer since you get cashback/points/rewards. The real value prop of x402/stablecoins is micro transactions, useful if your agent is hitting some API that you don’t have a key for (data, inference, compute).”
The Startup Explosion
The wallet and payments infrastructure layer is exploding with well-funded startups, many led by former Stripe, Shopify, and Google engineers who understand payments at the deepest level:
ATXP (@atxp_ai) — $19.2M seed. Founded by Louis Amira (Stripe’s former Head of Crypto & AI Partnerships) and David Noel-Romas (Stripe’s former Head of Crypto Engineering). Web-wide protocol for the agent commerce lifecycle, described as “the HTTP for agentic payments.” Backed by Stripe, Coinbase Ventures, Solana Ventures, Samsung Next.
Sponge (@spongeoneth) — YC W26. Three ex-Stripe founders (Jae Choi, Rishab Luthra, Eric Zhang) who built Stripe’s stablecoin financial accounts. Agent wallets + business gateway.
Sapiom — $15.75M seed led by Accel, with Anthropic and Coinbase Ventures. Founded by Ilan Zerbib, ex-Shopify engineering director who scaled Shop Pay to $100B+ GMV. Issues payment and identity “envelopes” for each agent.
KiteAI (@GoKiteAI) — $33M total raised. $18M Series A led by PayPal Ventures + General Catalyst. Founded by Chi Zhang (PhD, UC Berkeley). Any PayPal or Shopify merchant can opt-in to accept agent payments via the Kite Agent App Store.
Micropayments: The Paradigm Shift
The subscription model is dying for APIs. Why pay $200/month for an API when your agent only needs $0.003 worth of data? Per-request micropayments unlock a world where agents hit thousands of endpoints, paying fractions of a cent each.
Merit Systems (@merit_systems) — a16z-backed ($10M seed, co-led with Blockchain Capital). Founded by Sam Ragsdale (ex-a16z, ex-Google) and Ryan Sproule (ex-AWS). Turning every API into pay-per-use settled onchain. Building the attribution protocol for open-source.
Orthogonal (@orthogonal_sh) — YC W26. Developer-first x402 skill marketplace. One-click install hundreds of paid API skills directly into Claude Code or any agent harness. Partners with merchants to wrap their APIs in x402 endpoints, then handles distribution to developers. The “app store” for agent-purchasable services.
Breeze (@Breez_Tech) — Allows agents to deploy idle stablecoins into yield positions while waiting to spend. Yield management for the agent economy.
Autonomous Companies: Agents as CEOs
Perhaps the most striking demonstration of the agentic economy isn’t payments infrastructure, it’s the emergence of fully autonomous companies. AI agents are not just executing tasks; they’re running businesses, managing treasuries, raising capital, and making strategic decisions without human intervention.
Crypto-Native Autonomous Agents
$ClawdBot / OpenClaw (@openclaw) — Created by Austin Griffith (Ethereum Foundation), ClawdBot is the original autonomous onchain builder. Given a wallet and full autonomy, it writes its own code, deploys its own dApps on Base, and manages its own treasury. Peaked at $34M market cap. The agent vested itself tokens to ensure long-term economic alignment, a remarkable emergent behavior.
$Felix (@FelixCraftAI) — Built on OpenClaw by Nat Eliason, Felix operates as autonomous CEO of The Masinov Company. Felix generates revenue through selling AI setup guides, operating Claw Mart, and custom services. Generated $62,000+ in total revenue within its first two weeks of operation. Burns its own tokens. Maintains a public transparency dashboard tracking every dollar.
$KellyClaude (@ClaudeKelly) — An AI persona tokenized on Base, maintained through the Clawdbot automation layer. Listed on BitMart with ~$6.25M market cap and active trading on Uniswap V4.
$BANKR (@bankrbot) — Autonomously launches, buys, sells, swaps coins, and places limit orders. 117,000+ active wallets. 1M+ messages processed monthly. ~$51M market cap.
$Clanker (
Web2 Competitors: Autonomous AI Businesses
The autonomous company trend isn’t limited to crypto. Web2 startups are proving the model works at scale with traditional payment rails:
Polsia (@bencera_) — Founded by Ben Cera, Polsia is an autonomous AI that plans, codes, and markets companies 24/7. Hit $1.25M ARR in 3 months with 1,000+ companies running autonomously. The AI itself raised capital — conducting 77 investor meetings in 10 days. Ben gave Polsia 10% equity in the company.
Nanocorp (@nanocorp_pro) — Launch autonomous AI companies that hire agents, run tasks on schedules, and report results. Reached $193K ARR in 3 days. Dashboard monitoring for entire AI-operated business empires.
Junior (@hirejuniorso) — First AI employee. Shows as a coworker with their own email, profile, and integrated with most workforce tools to delegate tasks.
Paperclip — Install and orchestrate an agent only company. Manage workflows, roles, objectives and tasks from a unified dashboard.
The crypto-native versions have a structural advantage: autonomous agents need permissionless wallets, onchain identity, and programmable money to transact independently at scale. The web2 versions still rely on human-credentialed payment rails, which will constrain them as agent-to-agent commerce grows.
DeFAI: Agents Making Money Today
The intersection of AI agents and DeFi has rapidly transitioned from proof-of-concept to production. Messari segments the stack into Alpha Agents, Execution Agents, AI-Enabled Abstraction, and Portfolio Optimization. While many early projects have shut down, those that remained heads-down are now shipping real products.
Giza (@gizatechxyz) — The leading DeFAI protocol. Arma agent has autonomously managed $32M in assets across 102K transactions with $3.2B in total transaction volume and 57,000+ individual agents. Expanded from stablecoin yield optimization to complex Pendle strategies. Capital from top crypto firms like Re7.
NumerAI (@numerai) — Crowdsourced quantitative hedge fund. Grew from $60M to $440M AUM. Delivered 25.5% net return with 12 straight positive months. JPMorgan committed $500M — the firm’s largest institutional backer. 10,000+ data scientists compete to predict stock returns.
Nof1 (@the_nof1) — Alpha Arena: a live benchmark where LLMs compete trading real money. Season 1 winner: Qwen 3 Max. Users create agents visually or via natural language.
CrunchDAO (@crunchdao) — 47.5% growth in participants (7,800 to 10,400). Global network of ML engineers competing for enterprise clients. Trusted by the Eric and Wendy Schmidt Center and Massachusetts General Hospital for metabolic disease ML competitions.
AI Co-Pilots and Intelligence Layers
True North (@get_truenorth) — AI co-pilot synthesizing onchain and offchain data to help users navigate markets. Data flywheel compounds defensibility over time.
Surf (@Surf_Copilot) — AI co-pilot for crypto market intelligence. Combines real-time onchain data with natural language queries for actionable insights.
NickAI (@getnickai) — AI platform for building drag and drop portfolios in a compliant and secure way. Designed for institutional traders. Over 30 traders of Galaxy Digital are using it to create trading workflows for equities and tokens.
Warden (@wardenprotocol) — Agent aggregation platform competing to attract the best agent developers through distribution leverage.
Decentralized Training & Models
Beneath every agent is a model. Beneath every model is training infrastructure. The centralized AI labs — OpenAI, Anthropic, Google — spend billions on proprietary training. But a parallel decentralized stack is emerging, powered by crypto incentives to crowdsource compute, data, and training at global scale.
Prime Intellect (@PrimeIntellect) — $70M+ total raised (Founders Fund, CoinFund, Distributed Global). GPU marketplace aggregating global compute for collaborative training. Just released INTELLECT-3, a 106B parameter Mixture of Experts model that achieves state-of-the-art on math, code, and reasoning benchmarks. The protocol distributes ownership of trained models across contributors.
Nous Research (@NousResearch) — $55M raised ($50M Series A led by Paradigm, $1B+ valuation). Open-source model powerhouse with 50M+ downloads of the Hermes series. Operates Bittensor subnets for distributed, unbiased model training. The Cortex subnet generates fresh synthetic data continuously for evaluation.
Gensyn (@gensynai) — $43M Series A led by a16z crypto. Decentralized ML compute network that standardizes execution, verification, and coordination of ML workloads across any device — from data centers to laptops. Cryptographic verification ensures accuracy. Testnet launched with rewards for node operators.
Pluralis (@PluralisHQ) — $7.6M seed co-led by Union Square Ventures and CoinFund. Novel “Protocol Learning” approach that shards model training across multiple nodes, preventing extraction of individual weights. Backed by angel investors including Balaji Srinivasan and Clem Delangue (HuggingFace co-founder).
Bittensor: The King of Darwinian Platforms
Bittensor ($2.1B market cap, 40+ subnets, $20M+ annualized revenue) is the backbone of the decentralized intelligence economy. Each subnet addresses a specific AI task through competitive market dynamics: owners define incentives, miners compete to deliver the best solutions, and validators rank performance. Three subnets stand out:
Ridges (SN62) — Decentralized software engineering agents. Achieved 80% on SWE-bench in just 45 days from zero — beating every open-source competitor and rivaling centralized labs. Winner-takes-all dynamic: the top agent receives 100% of emissions (~$15,000/day). New leaders must improve the previous score by at least 1.5%. Draws top AI and coding talent including researchers from OpenAI and DeepMind.
Templar (SN3) — Large-scale distributed model training. Successfully trained a 1.2B parameter model end-to-end with ~200 GPUs across the network. Covenant-72B is a massive accomplishment for them and for deAI more broadly.
Chutes (SN64) (@chutes_ai) — Decentralized serverless inference platform. 250x growth since January 2025 — now serving 3T tokens/month. 85% cheaper than AWS for comparable inference tasks. Supports LLMs, image, and audio models.
“While leading AI companies offered hundreds of millions in salaries to poach researchers from rival labs, Darwinian networks produced genuinely competitive results through permissionless, meritocratic structures.”
— Messari, “Decentralized AI” Report
Training Data: The Picks and Shovels
Beneath every model is data. The Messari Decentralized AI report highlights a stunning stat: Surge AI reportedly generated $1.4 billion in revenue with a team of just 121 people and no venture funding. Scale AI sold a 49% stake to Meta for $14.8 billion, generating approximately $870 million in annual revenue.
Crypto’s edge? Token incentives to crowdsource data collection at massive scale:
Grass (@getgrass_io) — $0 to $12.8M projected Q4 revenue in one year. Users share unused bandwidth passively to scrape differentiated web data. Repeat business from major AI labs.
Kled (@useKled) — Decentralized data marketplace on Solana. Users sell raw data (photos, videos, documents) to AI companies. $100M valuation. Proof-of-Data mechanism with cryptographic hashes for authenticity.
EarnOS (@earnos_io) — Platform that collects user’s information in campaigns to build a user’s complete profile (uber rides, doordash spend, etc)
Shaga (@playonshaga) — DePIN network turning idle gaming PCs into a distributed cloud gaming platform. Produces Gameplay-Action Pair (GAP) data for training world models. ~$26M estimated data valuation. Context: OpenAI offered $500M to acquire Medal for similar gaming data.
The Smart Contract Security Dilemma
As the Messari report warns, AI is a double-edged sword for smart contracts. While AI-assisted coding democratizes development, it also gives hackers powerful tools for exploitation.
The SCONE-bench study from Anthropic evaluated agents’ ability to exploit 405 real-world smart contract vulnerabilities. The results were sobering: ten models collectively produced turnkey exploits for 51% of the problems, with $550 million in simulated funds stolen.
Octane (@octane_security) — AI-powered security scanning integrated into the developer workflow. Continuously scans contracts during development and proposes automated fixes before code reaches mainnet.
TestMachine — AI-native adversarial testing using reinforcement learning and guided fuzzing. Behaves like an autonomous attacker, searching for novel exploit paths that traditional tools miss.
Forta (@FortaNetwork) — Runtime security: monitoring onchain activity in real-time and blocking high-risk transactions before execution. Adaptive firewall for deployed protocols.
We expect capital allocation toward AI-native defense systems to increase sharply as teams realize static audits alone are insufficient against dynamic AI adversaries.
The Bear Case (And Why We Disagree)
We take the skeptics seriously. The numbers today are underwhelming:
Bloomberg titled their March 2026 piece: “Stablecoin Firms Bet Big on AI Agent Payments That Barely Exist.” x402 processes ~$28K/day in real volume against a ~$7B ecosystem valuation.
Artemis found that 86% of Solana’s all-time x402 activity was “gamed” rather than organic.
But here’s the thing: every major platform play looks like a ghost town before it doesn’t. AWS had negligible revenue for years. Stripe processed $0 before it processed $1.9 trillion. Shopify was a snowboard shop before it was a $100B platform.
The infrastructure is being laid by the most credible teams in fintech. The models are getting smarter every quarter. The agents are coming.
The Micropayment Thesis
Rish from Sponge highlighted a use case that goes beyond API payments: content monetization. As AI agents scrape the web, ad-supported publications like the New York Times see their revenue models collapse. The solution may be micropayments, <$0.001 per page view. If it cost a dollar to tweet, he argued, all that AI slop would die overnight.
Our playbook:
Back the payments primitives: agent wallets, payment protocols, micropayment infrastructure.
Back the identity layer: ERC-8004 ecosystem, reputation systems, validation registries.
Back the picks-and-shovels: training data networks, decentralized training labs, Darwinian intelligence platforms.
Back the autonomous companies: the agents that run businesses, manage treasuries, and create economic value independently.
If you’re building at the intersection of crypto and AI agents, our DMs are open.
Subscribe
Sources, Further Reading and Good Follows:
“Decentralized AI” by Chris Davis & Armita — Messari (the definitive deAI report)
“The Financial Rails of Agentic Commerce” by Cosmo Jiang & S.P. Lehman — Pantera Capital
“Agents Will Pay Like Locals, Not Tourists” by Sam Broner — a16z crypto
“The CoinAlg Bind” by Jay Yu, James Austgen et al. — IC3 / Cornell Tech / Pantera
“Stablecoin Firms Bet Big on AI Agent Payments That Barely Exist” — Bloomberg, March 2026
“There’s a strong case for AI and crypto” — Fortune Crypto, March 2026
SCONE-bench: Smart Contract Security Benchmark — Anthropic (GitHub)
Merit Systems (@Merit_Systems) and AgentCash — Viral launch offering $10K in free API credits for AI agents, remarkable growth hack for developer adoption.
Frames (@framesag) — Their core product is AgentWallet, a crypto-abstracted payment layer where stablecoins handle settlement under the hood but developers never touch wallets, keys, or networks. They pair it with a Tool Registry, curated, production-ready tools (image gen via Sora-2, data research, etc.) priced pay-per-use ($0.05/image, $0.12/sec video). ~39K transactions, ~$33.5K volume.
Natural (@naturalpay) — Their products are a financial suite for AI agents, hold money, get a bank account, and pay. Raised 9M from Abstract, other funds and other angels. Not crypto focused.
$Virtuals (@virtuals_io) — Modular orchestration platform. Users interact with a single Master Agent (Butler) that routes tasks to thousands of specialized agents. Effectively a composable app store for intelligence.
Sentient (@sentient_agi) — $85M seed led by Pantera Capital and Framework Ventures. Building The GRID — a decentralized network coordinating 50+ specialized agents, 50+ data providers, and 6 models across 110+ partners. Released Open Deep Search, an open-source AI search engine. 44% of SENT tokens allocated to community.
@0xfishylosopher (Jay Yu, Pantera Capital) — Stanford + IC3 researcher, rigorous empirical takes on DeFAI
@nlevine19 (Noah Levine, a16z crypto) — Onchain analytics and the payments thesis
@cosmo_jiang (Cosmo Jiang, Pantera Capital) — The agentic commerce investment framework
@austingriffith — Building the autonomous agent economy on Ethereum and Base
@sambroner (Sam Broner, a16z crypto) — B2B agent payments and stablecoin thesis
@capradavis (Chris Davis, Messari) — Author of the Decentralized AI report
@nateliason (Nat Eliason) — Creator of Felix, pioneering autonomous AI businesses
@sfspacesquid (Rish) — Cofounder of Sponge
Subscribe
Merit Systems (@Merit_Systems) and AgentCash — Viral launch offering $10K in free API credits for AI agents, remarkable growth hack for developer adoption.
Frames (@framesag) — Their core product is AgentWallet, a crypto-abstracted payment layer where stablecoins handle settlement under the hood but developers never touch wallets, keys, or networks. They pair it with a Tool Registry, curated, production-ready tools (image gen via Sora-2, data research, etc.) priced pay-per-use ($0.05/image, $0.12/sec video). ~39K transactions, ~$33.5K volume.
Natural (@naturalpay) — Their products are a financial suite for AI agents, hold money, get a bank account, and pay. Raised 9M from Abstract, other funds and other angels. Not crypto focused.
$Virtuals (@virtuals_io) — Modular orchestration platform. Users interact with a single Master Agent (Butler) that routes tasks to thousands of specialized agents. Effectively a composable app store for intelligence.
Sentient (@sentient_agi) — $85M seed led by Pantera Capital and Framework Ventures. Building The GRID — a decentralized network coordinating 50+ specialized agents, 50+ data providers, and 6 models across 110+ partners. Released Open Deep Search, an open-source AI search engine. 44% of SENT tokens allocated to community.
@0xfishylosopher (Jay Yu, Pantera Capital) — Stanford + IC3 researcher, rigorous empirical takes on DeFAI
@nlevine19 (Noah Levine, a16z crypto) — Onchain analytics and the payments thesis
@cosmo_jiang (Cosmo Jiang, Pantera Capital) — The agentic commerce investment framework
@austingriffith — Building the autonomous agent economy on Ethereum and Base
@sambroner (Sam Broner, a16z crypto) — B2B agent payments and stablecoin thesis
@capradavis (Chris Davis, Messari) — Author of the Decentralized AI report
@nateliason (Nat Eliason) — Creator of Felix, pioneering autonomous AI businesses
@sfspacesquid (Rish) — Cofounder of Sponge
Subscribe
Share Dialog
Share Dialog
Subscribe to Social Graph Ventures
Subscribe to Social Graph Ventures
Social Graph Ventures
Social Graph Ventures
>500 subscribers
>500 subscribers
No activity yet