bet

DeFi has long promised open and self-custodial finance. But for most users, actually using it still means juggling through wallets, dApps, bridges, pools, approvals, and risks that are very hard to understand in real time, especially for someone who’s relatively new to the industry.

CoinFello believes that the experience is ready for a major shift. With Fello 1, the company is building a self-sovereign AI agent designed to help users interact with DeFi through plain language while keeping complete control over their wallets and keys.

In the following interview with the founder, we go through why agents could become the primary interface for onchain finance, how controlled delegation can make automation a lot safer, and why liquidity provision is one of the first major frontiers for agent-powered decentralized finance.

CoinFello is positioning itself as a self-sovereign AI agent for DeFi. In simple terms, what problem are you trying to solve that wallets and dapps have not solved yet?

CoinFello is a completely new way to understand, use, and automate smart contracts.

The previous paradigm required users to create a wallet, navigate many disjointed websites, connect that wallet to a website, and then almost blindly trust that the smart contracts on that website do what the website promises they do. This made DeFi inaccessible, extremely complicated, and dangerous, and was one of the primary barriers to broader DeFi adoption.

CoinFello’s approach is to give users an agent that can interface directly with the smart contracts through a Claude-like user experience that people are familiar with. The agent isn’t just easier to use, it also opens up new frontiers of automation, where agents can act on behalf of users to accomplish virtually anything in DeFi: batch swap multiple tokens and bridge them across networks, discover advanced yield strategies, optimize existing deposits, take out a loan and automate payments, and a whole lot more. CoinFello makes doing these things super simple.

Fello 1 is described as a general-purpose DeFi agent rather than a narrowly integrated assistant. Why is general-purpose execution important, and what does it unlock for users that protocol-specific interfaces cannot?

DeFi is not one app or use case.

DeFi is an ecosystem of contracts, protocols, pools, vaults, bridges, and networks that constantly change.

Unfortunately, most of the crypto AI agent products on the market are just trading bots connected to some centralized API. If an agent only works through narrow integrations, it will always be limited to a few narrow use cases. That’s not how people use the internet (web browsers), their phones (extensible smartphones), AI agents, or even Ethereum itself. All of the great innovations were fundamentally extensible.

General-purpose execution means Fello 1 can reason about and interact with EVM-compatible smart contracts more broadly, instead of being locked into a small set of pre-built workflows. That unlocks all kinds of use cases that we ourselves never anticipate or integrate with. New pools, new protocols, and new opportunities can become accessible faster, without waiting for a dedicated front end or a code release for every specific action.

For the user, the benefit is simple: they do not need to jump between ten different interfaces to complete one DeFi strategy. They can describe what they want, review the steps, and execute across protocols from one agentic interface.

One of CoinFello’s core promises is that users can interact with DeFi through plain language while keeping custody of their wallets and private keys. How do you balance ease of use with the security expectations of self-custody?

We’ve tried to bring self-custody principles to the agentic era. This means that funds must remain in a self-custodied wallet, and agents should have guardrails enforced on them that define what funds they can access, in what ways those funds can be used, and for how long that agent has access to those funds.

With Fello 1, users keep their wallets and private keys. The agent operates through limited permissions that the user chooses to grant, and users review and approve transactions before execution. Plain language is the interface layer, not a replacement for consent. We fundamentally disagree with the approach of transferring funds to a centralized trading bot and hoping for the best.

The goal is to reduce cognitive overload without reducing user sovereignty. Fello can do the math, explain the route, surface the risks, prepare the transaction, and monitor positions, but the user remains in control of what permissions exist and what actually gets executed.

The Fello 1 launch puts a lot of emphasis on liquidity provision, including Uniswap V2, V3, and V4 positions, fee tiers, impermanent loss, and live position monitoring. Why did you choose LP management as such an important use case for the product?

Liquidity provision is one of the best examples of DeFi’s promise and its complexity. Concentrated liquidity can be a powerful yield opportunity, but it asks a lot from the user. You need to understand price ranges, ticks, fee tiers, pool selection, position sizing, impermanent loss, and when your liquidity is in or out of range.

That is exactly the kind of experience where an AI agent can create real value. Fello 1 can handle the mechanical and analytical parts: identifying LP strategies, doing the math, monitoring the position, explaining whether it is in range, showing the real return, and helping the user understand the trade-offs.

We chose LP management because it is not just a button-clicking problem. It is a decision-support problem. If we can make LPing understandable and manageable for more users while keeping them self-custodial, that is a major step toward making DeFi more mainstream.

AI agents in crypto are often associated with automation, but CoinFello says Fello 1 is not designed as an autonomous trading bot and that users still review and approve transactions. Where do you draw the line between helpful automation and too much delegation?

To be clear, we are building for automation, and we deeply believe users should be able to delegate approval for tightly defined automations to their agent. These are very complex problems to solve, so we’ve been working to expand the agent’s capabilities and the kinds of automation the user can create through the permissions and delegations we’ve been championing.

You previously led operations at MetaMask, one of the most important wallet products in crypto. What did that experience teach you about user behavior, wallet UX, and self-custody that directly shaped CoinFello?

MetaMask had a very radical vision in the early days of Ethereum. Most people at the time were building “use case wallets” with a handful of brittle integrations. MetaMask sought to do something else: create a permissionless and extensible wallet that could be used with any smart contract protocol.

We’ve brought the same radical values and vision to CoinFello that we previously used to build MetaMask. While most in the agent space are building narrow “use case bots,” our goal is different: to bring users onchain, and give them access to the entire decentralized web.

We also learned about the limitations of trying to solve the safety and user experience problems at the wallet layer. Wallets are forced to maintain endless integrations with third party protocols, and these integrations make their products slow to innovate, highly prone to bugs, and generally dangerous because the wallet still can’t understand what a smart contract *actually does.*

CoinFello is how we will solve these problems for the next wave of on-chain innovation.

CoinFello relies on a delegation model where users grant agents limited permissions that can be modified or revoked. What does a safe permission system for onchain AI agents need to look like as these tools become more powerful?

A safe permission system needs to be specific, limited, transparent, and revocable.

Users should not have to grant broad, unlimited authority over funds to an agent. Permissions should be scoped by action type, asset, protocol, amount, duration, and any other relevant rule the user cares about. The user should be able to see what permissions exist, understand what they allow, and revoke or modify them at any time.

As agents become more powerful, permission design becomes one of the most important parts of the stack. The future is not giving the AI your keys. The future is controlled delegation, where the agent can help execute within boundaries that the user defines. That is how we get the benefits of automation without sacrificing self-sovereignty.

Looking ahead, do you think the future of DeFi will still be built around users manually navigating dapps, or will agents become the primary interface for onchain finance?

I think dapps will still matter, but agents will become the primary interface for most users.

Today, DeFi still looks like the early internet in some ways. Users manually navigate different websites, learn different interfaces, and stitch together actions themselves. That works for power users, but it does not scale to broader adoption.

Agents change the interface from navigation to intent. Instead of asking users to know exactly which protocol to use and which buttons to click, they can say what they want to accomplish, compare options, understand risks, and approve execution.

The future of on-chain finance will still be open, composable, and self-custodial. But the way users access it will become much more conversational, automated, and personalized. Our view is that agents will become the execution layer that makes DeFi usable for the next wave of users.

Disclaimer: The content shared in this interview is for informational purposes only and does not constitute financial advice, investment recommendation, or endorsement of any project, protocol, or asset. The cryptocurrency space involves risk and volatility. Readers are encouraged to conduct their own research and consult with qualified professionals before making any financial decisions. This interview was conducted in cooperation with CoinFello, who generously shared their time and insights. The content has been reviewed and approved for publication in mutual understanding. Minor edits have been made for clarity and readability, while preserving the substance and tone of the original conversation.