The dominant narrative in crypto goes something like this: blockchain needs AI to finally achieve mainstream adoption. AI will make smart contracts smarter, DeFi more accessible, and on-chain experiences more intuitive. It's a comforting story. It's also backwards.
The real insight is that AI agents need blockchain -- not as a feature, but as a requirement. Autonomous software that transacts, negotiates, and allocates resources cannot operate on infrastructure designed for humans with bank accounts and legal identities. It needs permissionless rails. And the infrastructure gap between what agents need and what traditional finance offers is not a problem that incremental API improvements will solve.
The Permissionless Requirement
Consider what an autonomous agent actually needs to function as an economic actor. It needs to hold assets, execute transactions, enter agreements, and prove its actions -- all without a human co-signer. Traditional financial infrastructure doesn't support this. Every bank account requires a human. Every API key can be revoked. Every payment processor has terms of service written for people.
Blockchain provides the only infrastructure where software can operate as a first-class economic participant. Permissionless access means no gatekeeper can prevent an agent from transacting. Programmable money means agents can encode complex economic logic directly into their transactions. Cryptographic verification means every action is provable. Composability means an agent can build on top of any existing protocol without negotiating a partnership or signing a contract. These properties aren't just nice-to-have for agents -- they're table stakes for any software that needs to operate autonomously in an economic context.
The contrast is stark when you look at specifics. Uniswap processes billions in volume with no API keys, no approval process, no KYC -- just a smart contract that anyone (or anything) can call. An agent can provide liquidity, execute swaps, and collect fees without ever proving it's human. Try doing that through Stripe or Plaid. Those systems require human identity verification, and their access is revocable at any time. Safe (formerly Gnosis Safe) already enables multi-sig wallets that agents can operate as signers. ERC-4337 account abstraction takes this further, enabling smart contract wallets purpose-built for non-human operators -- wallets that can enforce spending limits, require multi-agent approval, and execute complex transaction logic without a private key held by a person.
The Agent Stack Is Emerging
What makes this moment different from previous "crypto + AI" cycles is that a genuine infrastructure stack is forming -- not as a single project, but as a constellation of protocols that, assembled correctly, give agents everything they need to operate autonomously.
For execution, Flashbots' SUAVE is building an intent-based transaction system where agents express what they want to achieve, and a decentralized network of executors competes to fulfill those intents optimally. This is fundamentally different from agents submitting raw transactions -- it's an execution layer designed for machine participants from the ground up.
For identity, ENS provides human-readable naming, but the more interesting primitive is ERC-6551 -- token-bound accounts that turn NFTs into wallets. An agent's identity becomes an on-chain entity that owns assets, accumulates history, and carries reputation. Combine this with attestation services and you get the beginning of a machine credential system.
For compute, Ritual is building infrastructure for on-chain inference requests, while Giza enables zero-knowledge proofs of machine learning model execution -- letting agents prove what model they ran and what output it produced without revealing the model itself. This is critical for any agent that needs its decision-making to be auditable by counterparties or governance systems.
For payments, Circle's Cross-Chain Transfer Protocol enables stablecoin movement across chains, and Superfluid allows streaming payments -- an agent paying another agent per-second for an ongoing service, settled on-chain without invoices or reconciliation. This streaming payment model is particularly native to agent economics, where services are continuous rather than discrete and billing should be granular rather than monthly.
The stack exists in pieces. The venture opportunity is in the teams assembling these pieces into coherent platforms -- and in the missing layers that haven't been built yet. Agent-to-agent negotiation protocols, decentralized reputation systems for software entities, on-chain dispute resolution for autonomous transactions -- these are the gaps that define the next generation of infrastructure companies.
What's notable is the pace. Two years ago, none of these projects existed in production. Today, each layer has multiple teams shipping code. The infrastructure isn't hypothetical -- it's being built in real time, and the teams that move fastest to integrate across layers will have a compounding advantage as agent adoption accelerates.
Beyond the Wrapper Thesis
Most "AI + crypto" projects today are wrappers -- they put a chatbot on top of a DEX or use an LLM to generate smart contract code. These are features, not paradigm shifts. Look at ChatGPT plugins or GPT-4's browsing capability: centralized agent patterns where access is revocable, the agent has no economic agency of its own, and nothing it does is composable with other systems. OpenAI can shut off any plugin at any time. The agent owns nothing, controls nothing, and builds no persistent state. The same limitation applies to every centralized agent framework: the agent is a tenant on someone else's platform, subject to that platform's rules, pricing, and continued existence.
The venture-scale opportunity isn't making blockchain more user-friendly. It's building the blockchain-native infrastructure that agents require to operate autonomously. Protocols designed from the ground up for machine interaction: settlement layers optimized for microsecond execution, identity systems that work for software entities, and governance mechanisms where agents can participate as stakeholders. On-chain agents own assets, execute autonomously, and compose with every other protocol in the ecosystem without asking permission.
What We're Backing
At Isoline, we invest in the infrastructure layer -- the protocols that will be invisible to end users but essential to every agent. We're focused on three categories: settlement infrastructure that handles agent-to-agent transactions at the speed and cost agents require, agent wallet systems that give software entities full economic agency, and machine-readable protocols that are designed for programmatic interaction rather than human interfaces.
We're not alone in seeing this. Coinbase launched an agent toolkit for on-chain transactions. a16z is backing agent infrastructure across their crypto portfolio. The major crypto funds are quietly repositioning around the thesis that agents are the next major user base for on-chain protocols. But most are still thinking about this as "AI features for crypto products" rather than "crypto infrastructure for AI agents." That framing difference is where we see our edge.
The market will eventually converge on the realization that agents are not a feature of the crypto economy -- they are its next native user base, larger in transaction volume and more demanding in infrastructure requirements than human users ever were. The protocols that capture this shift early will define the next era of on-chain activity. We want to be on their cap tables when that happens.
The question isn't whether AI will use blockchain. The question is which protocols will become the default rails for agent commerce -- and we intend to back them before the market catches up.