CoinFello COO on building the execution layer for autonomous onchain gents

We spoke with MinChi Park, Co-founder and Chief Operating Officer of CoinFello, about their go-to-market strategy for agent infrastructure and how her team is developing the execution layer for autonomous onchain agents.

Read more about how CoinFello is transforming the crypto user experience from manual clicks to delegated automation in the interview below.

You came up through partnerships and ecosystem roles at HyperPlay and Windranger before co-founding CoinFello. How did that background shape how you think about the go-to-market side of building agent infrastructure?

Coming up through ecosystem and partnership roles fundamentally shaped my perspective on distribution. At HyperPlay and Windranger, the core focus was building the relationships and developer surface area necessary to make our features successful.

You constantly ask who needs to believe in the vision before users will, which forces a deep focus on builders, protocols, and communities. This background also trained me to operate at multiple levels of the market simultaneously.

At close range, you track individual partners, observing the evolution of MetaMask’s smart accounts strategy and the actual priorities of the Ethereum Foundation’s dAI team.

Simultaneously, you maintain a bird’s eye view of the essential protocols, the best collaborations to pursue, and where standards are solidifying. Ecosystem roles force you to develop this multi-level reading as a core survival skill. That wired me to architect GTM directly into the product from day one.

ERC-8004 is a prime example. Having been in close contact with the Ethereum Foundation’s dAI team since late last year, we deliberately aligned with this agent registry standard early, well before widespread adoption or mandatory compatibility. Building on 8004 alongside ERC-7710 was a calculated bet on the future of these standards.

The same principle applies to developing the skill as an open-source, MIT-licensed project on ClawHub. These technical choices serve a dual purpose: they make us understandable to the relevant communities and reduce obstacles to adoption. Ecosystem work also teaches you how to manage urgency mismatches.

Enterprise cycles and protocol governance operate over long timelines, while consumer behaviour changes rapidly. You become better at recognising the moments when those timelines align and moving quickly into those opportunities. The OpenClaw viral moment in January was one of those moments, and we seized it because we were already monitoring for it.

CoinFello is described as a “white glove crypto agent.” What does that mean for a user’s day-to-day experience versus just using a DEX or a wallet?

A DEX provides a swap interface, and a wallet displays your balances. Both assume you know what you want to do, how to do it, and that you’ll sit there and click confirm. While that approach works for some users, most people prefer a smoother interaction with their finances.

In the world of blockchain, the terminology alone often causes people to revert to a centralised exchange. White-glove means the agent manages the entire process with the extra care the user deserves. You say “stake 0.1 ETH on Base”, and it figures out the staking protocol, the gas settings, and the contract approvals.

It constructs the transaction, routes it correctly, requests only the required delegation, and returns a block explorer link upon completion. The underlying infrastructure handles much of the work, making this process appear simple. ERC-4337 smart accounts, ERC-7710 fine-grained delegations, and CoinFello’s natural language to the transaction routing layer all operate entirely in the background.

The user simply sees the agent ask for permission to spend 0.1 ETH, says “yes,” and the action completes. That is the experience we optimise for.

BuffiBot helped ETHDenver attendees navigate hundreds of sessions through natural language and voice. What did you learn from running it live, and how did that shape where the core product is headed?

ETHDenver was a massive milestone for us. Putting the product in front of thousands of people in noisy rooms with weak WiFi gave us an incredible, immediate feedback loop. You quickly learn that people use tools exactly how they want to, regardless of your original design.

BuffiBot proved that a natural language interface instantly clicks with people. Users experienced zero comprehension gap. One of the coolest takeaways was seeing attendees naturally switch to their native languages. The bot just worked perfectly in whatever language they were thinking in at the time. We are bringing that exact level of intuitive tailoring to CoinFello.

The big “aha” moment at ETHDenver happened when attendees simply typed, “I want to mint the official POAP.” BuffiBot handled the entire mint right there in the chat. The user stayed completely in flow, without ever needing a separate page or a copied wallet address. The agent just executed the task seamlessly. We are collapsing all onchain actions into that conversational interface for CoinFello.

Developer interest was another huge signal. Teams at ETHDenver immediately wanted their own product agents to access CoinFello. That enthusiasm heavily accelerated our focus on the agent-to-agent protocol and the ERC-8004 registry. We realised our market includes both everyday users and other builders wanting to integrate smooth agent execution.

CoinFello analyses wallet history and interprets user intent before surfacing protocols and tokens. How does that personalisation layer work?

It starts by analysing the wallet’s contents, including tokens, chains, and previously used protocols. This gives CoinFello a strong baseline context. When a user sends a prompt, we route it directly against their actual on-chain profile to ensure immediate relevance.

We rely on natural language processing to interpret intent. A prompt like “Move some of my stables” is inherently vague. CoinFello resolves that ambiguity by checking the wallet context, looking at held stablecoins, past movements, and current gas conditions.

The goal is for the agent to proceed smoothly, ideally asking for clarification only once to confirm the scope of the delegation. For example, if you say “move some of my stables” and you only hold USDC on Base, the agent already understands the context and recognises your intent automatically.

Your wallet history does the heavy lifting, saving you from having to manually specify the token, chain, or contract. This personalisation compounds in value over time and provides significant protection. With our automation and delegation features, CoinFello keeps users safe during Black Swan events that occur while they sleep.

Look at October 10th last year, when the market was overleveraged following major tariff announcements. Over $19 billion in positions were liquidated within hours, wiping out 1.6 million traders. Exchange interfaces froze, and traditional stop-losses failed completely. Users holding open Aave loans were liquidated in the cascade before they could react.

With CoinFello, you can instruct the agent to monitor your loan positions and take protective actions if liquidation is imminent. You delegate authority, the agent keeps an eye on the threshold, and it acts automatically. That is the true, practical value of delegated automation.

The CoinFello skill on ClawHub lets any OpenClaw MoltBot install onchain execution in a single command. What was the thinking behind plugging into that ecosystem specifically?

The CoinFello skill on ClawHub lets any OpenClaw MoltBot install onchain execution in a single command. What was the thinking behind plugging into that ecosystem specifically?

OpenClaw went viral simply by solving a real problem. It provided a highly accessible personal agent for inboxes, calendars, and flights. The massive spike in GitHub stars, forks, and downloads in just 30 days proved its value. The user base is growing rapidly with people who prefer running applications locally, demand absolute control, and feel right at home in a terminal.

Those are our exact target users. The skills model provides the perfect technical architecture for what we are building. CoinFello functions purely as the execution layer for onchain actions. By using skills, we can cleanly extend an agent’s capabilities while leaving core runtime development to others.

ClawHub acts as a seamless distribution point. A user can go from zero to staking ETH in under five minutes, completely bypassing the hassle of separate accounts or new interfaces. We built this skill to the agentskills.io specification, ensuring it integrates perfectly with Claude Code alongside OpenClaw.

We are positioning CoinFello as the universal crypto execution layer for the entire agent ecosystem. We made it MIT-licensed and open source so anyone can audit, fork, and build upon it. In a space where trust is the actual product, building completely verifiable systems is mandatory.

How do you design granular delegation in a way that is both understandable and useful to a user who is not deeply technical?

The fundamental design principle dictates that users never need to understand ERC-7710 or ERC-7715 to use them. These standards allow developers to implement precise, revocable, and programmable permissions. ERC-7710 handles the delegation framework, covering actors and scope, while ERC-7715 enables advanced conditions, time limits, and spending caps.

They form the technical foundation for all our permission layers. The user avoids this complexity, simply seeing a prompt requesting permission to manage 0.1 ETH, which they easily understand and confirm. The complex work occurs in the parsing layer, translating a natural-language prompt into a highly specific delegation scope.

A command to stake 0.1 ETH produces a delegation precisely scoped to that action, chain, and amount, preventing total wallet access. If an issue occurs through prompt injection or a malicious skill, the blast radius remains strictly limited to the explicitly authorised amount.

CoinFello’s ERC-8004 construction enables other agents to discover and call it onchain algorithmically. What kinds of agent-to-agent interactions are you seeing or anticipating?

The immediate use case functions as an execution handover. A general-purpose agent receives a financial instruction it cannot fulfil and delegates it directly to CoinFello. The agent-to-agent protocol manages structured communication, while ERC-8004 handles verifiable discovery. The calling agent only needs to know that CoinFello can manage the task, without needing to understand its internal processes.

A more advanced use case involves agent economies managed through programming. An agent overseeing a project budget needs to pay other agents for services like model inference, data access, and research. CoinFello’s delegation model, combined with the ERC-8004 registry and the x402 monetisation protocol we are developing, offers the precise infrastructure for this.

It makes agent-to-agent payment infrastructure function at scale by automating actual smart contracts on-chain. At ETHDenver, many builders wanted to introduce CoinFello to their own products’ agents. We observed demand from portfolio trackers for rebalancing features and from lending products for automatic position closures.

These are B2B agent integrations rather than direct end-user interactions, and we are actively developing for this market alongside the consumer experience.

The broader AI agent infrastructure race is accelerating: Coinbase with agentic wallets, Olas with 4 million agent transactions, Lightning Labs with L402. Where does CoinFello fit within that ecosystem?

The current industry race focuses heavily on custody infrastructure and providing agents with their own wallets. Providers like Coinbase, Privy, and Dynamic are addressing various aspects of that specific challenge. At CoinFello, we focus on a distinct set of operational and security challenges.

The traditional wallet-per-agent model creates a bottleneck for the user. Each new agent requires a dedicated wallet and upfront funding before it can perform actions. This causes significant operational friction at scale. We designed CoinFello’s delegation model to eliminate this friction entirely. The agent operates from a scoped permission linked directly to your existing wallet.

Users can skip creating new wallets, pre-loading funds, and managing separate balances. We focus entirely on expanding the agent’s capabilities, its precision, and its safety parameters. While Trusted Execution Environments improve security for agent wallets, those setups remain custodial.

If a provider’s infrastructure goes down, the agent’s wallet goes down with it. CoinFello is engineered purely for self-custody and strict delegation. Your keys stay with you. You grant a specific, scoped permission to CoinFello, ensuring every step remains verifiable, explicitly consented to, and easy to revoke. This establishes a fundamentally distinct security model.

If the autonomous agent thesis plays out, what will the typical DeFi user look like in three years, and will there still be a “user” in the traditional sense?

The traditional DeFi experience of manually checking dashboards and approving single transactions will become a much smaller piece of the pie. People will still be the driving force. Agents will simply take over the routine execution, freeing up humans to focus entirely on high-level strategic decisions.

Ownership and clear intent remain the core of the system. A user will decide they want a specific percentage of their portfolio in staked ETH and set that goal. The agent then takes over the execution, monitoring, and adjusting of that position. Humans will operate at a much higher level of abstraction, providing final approval for any major changes or meaningful thresholds.

We are going to see a massive shift in protocol design as agents become the primary interface. Protocols must begin exposing clean, machine-parseable APIs to accommodate this shift. Operations that currently take ten manual steps will need to be reduced to a single execution.

Infrastructure builders focusing on agent-first consumption through standards like ERC-4337 and structured delegation will secure a massive structural advantage. We are building our entire platform for that exact future.