Somebody at CoinFello decided the interesting question was not whether an AI agent can execute a swap. That was settled a year ago. The interesting question is whether anyone will let one run for a month without being asked again.

Fello 2, released Wednesday, is the company's answer, write a strategy the way you would explain it to a colleague, approve the plan the agent writes back and then stop thinking about it. The example CoinFello leads with is not a command but a set of conditions: move idle USDC into the best stablecoin pool every Friday, but if a position drifts within 15% of liquidation, deal with the margin first. Three instructions, a dependency between two of them, a reporting trigger at the end.

Fello 1, which arrived in June, could already handle the hard parts of that in isolation. It did tick math on Uniswap V2, V3 and V4 positions, tracked whether a range was live and priced impermanent loss in dollars rather than leaving the user to work it out. What it could not do was remember. The conversation ended and so did the strategy.

Alongside the agent, CoinFello turned off its own revenue.

The fee news is the real news in this competitive web3 market

Every eligible swap through the product earns the trader a cut of the fee it generates. Refer somebody and you earn from their trading too, for as long as they keep trading. During the launch period CoinFello keeps none of it.

Zero is not new. Uniswap's wallet went to zero in late 2025 and Rabby has charged a quarter point for years. What CoinFello has done is put itself level with the floor rather than above it, which only matters because of where its likely users are coming from. MetaMask charges 0.875% and has something like 100 million installs behind it. That is the wallet CoinFello is trying to peel users away from. Against that number, a full rebate is a real argument.

Whether the argument is worth anything to you depends on how much you actually trade. Somebody moving a thousand dollars a month is being handed about a hundred dollars a year. Somebody running fifty thousand a month through MetaMask is being handed five thousand. The rebate is not really a feature. It is a targeting mechanism. The people it catches hardest are the ones already trading enough to have noticed what they were paying.

The reward is a token, not a rebate

Rewards settle in MON, which means the value of a rebate earned on Monday is whatever MON is worth on the day it gets claimed. Monad burns the base component of every transaction fee and mints new MON to stakers, so the asset has its own supply dynamics that have nothing to do with how much anyone swapped.

A rebate denominated in a volatile asset is a different product from one denominated in USDC. A referral program built on it is a different pitch again. The chain underneath is at least moving in the right direction for what CoinFello is trying to do.

Monad cut its consensus interval from 400ms to 300ms in July, taking deterministic finality from 800ms to 600ms. For a product whose whole premise is conditional execution, that number is not decoration. An agent told to add margin before a position crosses a liquidation threshold is racing the same market everyone else is racing. The window it has to act in is bounded by how fast the chain settles.

The trade-off is worth knowing. The shorter interval came with a lower per-block gas limit, cut from 200 million to 150 million. Faster blocks, less room in each one.

The claim that has to hold

CoinFello says Fello 2 inspects what it is asked to sign. It declines blind-signing requests, refuses approval-phishing attempts and rejects transactions built to empty a wallet. It says the agent will never take permissions broader than the ones granted.

None of that is verified anywhere in the launch materials. No audit, no methodology, no statement of what the agent does when it meets a contract it cannot classify. For software that executes on a Friday schedule while its user is asleep, this is the claim everything else rests on. It is also the only one with nothing underneath it. The delegation model is a better answer than the promise. Spending caps, allow-lists and expiry dates limit what a mistake can cost by construction rather than by good behaviour. That is architecture rather than assurance. It is why the product holds up even if the phishing detection turns out to be ordinary.

The founder's background explains the choice. jacobc.eth ran operations at MetaMask before this and was head of product at Mantle. Fello 1 was built on MetaMask's own smart account standards. CoinFello did not go off and invent a custody model, because the person running it had watched what happens to companies that do.

Shipping fast, pricing later

The company first surfaced in November, opened to the public in March and has shipped two major agent versions since. The gap between releases has fallen from 133 days to 62. Put that next to a total fee giveaway and an open fundraise and the strategy is not hard to read: take the ground now, work out the rent later. Land grabs are a reasonable strategy. They also have a well-documented failure mode, which is that users acquired at zero leave at zero.

The launch materials describe CoinFello as the first agentic product for using and automating all of DeFi. Its own earlier announcements were more careful. In November it claimed first self-sovereign agent, which is a position it can hold. Somewhere between then and now the qualifier fell off. Without it the claim does not survive a single search.

Giza's ARMA has spun up more than 25,000 agent instances managing over $35M of user money across 102,000 transactions, non-custodially, using session keys and smart accounts. Almanak runs an eighteen-agent swarm around a stablecoin vault. Brahma, HeyAnon and Wayfinder sit in what The Block calls the second wave of transactional agents.

CoinFello does have a real distinction. ARMA optimises stablecoin yield and does it well; Fello 2 will attempt an arbitrary EVM contract. Breadth against depth is a genuine argument and the company did not need to overclaim to make it.

What to Watch

The rate is the whole thing. CoinFello has promised to publish its standard split before the free period closes. That number will say more about the business than Fello 2 does. A generous split, something a user can still feel at ordinary volume, means the company intends to make its money on automation and treat swaps as the on-ramp. A thin one means the rebate was an acquisition cost and the product is a wallet with better manners. Watch for the number, then watch how much notice comes with it.

The second thing to watch is quieter. EIP-7702 made scoped delegation available to every wallet on the EVM. Agent builders have no special claim on it. The pattern Fello 2 depends on is not proprietary. MetaMask could ship it to a hundred million people without acquiring a single new user. The man who used to run its operations knows that better than anybody.

Don’t forget to like and share the story!