UniswapX and CoW Swap: How Intents Proved Out On-Chain

From Mempool to Outcome: What Changed

Imagine sending 5 ETH to a DEX and watching a bot sandwich your trade in the same block — your expected $14,800 USDC lands at $14,560. This is not a bug. It is the predictable outcome of submitting a transaction to a public mempool with a fixed execution path and paying gas upfront.

Intent-based protocols solve this differently. Instead of constructing the execution path yourself, you sign an off-chain order that specifies an outcome: "I will give you my 5 ETH; I require at least 14,600 USDC in return, deadline in 90 seconds." A competitive layer of professional intermediaries — called fillers on UniswapX and solvers on CoW Swap — races to fill that order at the best price they can source. They pay the gas. You receive the proceeds.

The orders use EIP-712 typed structured-data signatures, binding each order cryptographically to its exact parameters: input token, output minimum, deadline, and a nonce preventing replay. Because orders are submitted off-chain and never enter the public mempool, the primary MEV (maximal extractable value) attack surface disappears before the auction even begins.

Part 1 of this series gave you the mental model — intent as an outcome specification rather than a transaction. This article shows how two different auction designs made that model real on Ethereum mainnet.

UniswapX: Individual Dutch Auctions

Uniswap Labs launched UniswapX in beta in 2023 on Ethereum mainnet, later expanding to additional chains. Its core mechanism is a per-order Dutch auction.

When you submit an order, the protocol sets a starting output — slightly above your stated minimum — and linearly decays it over a typical 30-to-60 second window down to your floor. Any filler watching the auction calculates their cost to source the required output token. The moment the decaying auction price exceeds that cost, filling becomes profitable. Competitive pressure means fills tend to happen early in the window, delivering a price above the floor.

Two-tier execution adds speed for liquid pairs. UniswapX first attempts an RFQ (Request for Quote) fill: a private, near-instant quote from a whitelisted market maker. If no RFQ is taken, the Dutch auction opens publicly to any filler that implements the required interface.

Fillers are permissionless in principle. In practice, the competitive field consists of professional market makers, MEV searchers, and arbitrage bots — parties with the capital and infrastructure to source liquidity efficiently. They price gas costs into their spread, typically one to five basis points on major pairs, and can batch multiple fills into a single transaction to amortise that gas across orders.

Token transfers use Permit2's SignatureTransfer mode — one-shot and non-replayable. The signed order grants spending authority only for that single fill, leaving no persistent allowance afterward. (The initial Permit2 approval still requires one on-chain transaction and gas.)

Reported performance has been strong: roughly 80% of UniswapX swaps have settled at a better price than the user's initial quote, with MEV that would have gone to sandwich attackers redirected back as price improvement instead.

In 2024, UniswapX introduced cross-chain order support conforming to ERC-7683 — co-authored by Uniswap Labs and Across and introduced in 2024 — but single-chain Dutch auction orders remain the dominant use case.

CoW Swap: Batch Auctions and Coincidence of Wants

CoW Protocol takes the opposite architectural approach. Rather than settling each order immediately, it collects orders over a discrete time window and settles them together in a single on-chain batch auction.

Solvers — bonded and DAO-approved professional parties — compete to propose the optimal settlement for each batch. The solver that delivers the greatest aggregate surplus (total user value above their stated minimums) wins the right to execute and earns COW token rewards. Misconduct can be punished through bond slashing, keeping solver incentives aligned with users.

The batch model has two structural advantages:

Uniform Clearing Price (UCP): All orders trading the same token pair within a batch clear at the same price. Because every order settles simultaneously, transaction reordering inside the batch is irrelevant — a sandwich attacker cannot insert a transaction between two orders that clear together.

Coincidence of Wants (CoW): If one user wants ETH→USDC and another wants USDC→ETH in the same batch, a solver can match them peer-to-peer at the clearing price without routing either order through an automated market maker. No pool fees, no price impact. Both users save; the solver captures a smaller spread. When a full CoW match is not possible, solvers route remaining volume through on-chain AMMs (Uniswap, Balancer, Curve) or off-chain inventory.

CoW Protocol processed $40.2 billion in volume in 2024 and $87 billion in 2025 — its largest year — with monthly volume first exceeding $9 billion in July 2025 and sustaining above $10 billion per month on Ethereum in late 2025. By early 2026 the protocol had expanded to Arbitrum, Base, Gnosis Chain, Avalanche, Polygon, and Ink (Kraken's L2).

CoW Hooks extend composability further: users can attach arbitrary actions before a swap (pre-hooks) or after it (post-hooks) — unstaking tokens before a sell, depositing proceeds into a lending vault after — all atomically within the same batch transaction.

Head-to-Head: Dutch Auction vs Batch Auction

Both require only a single token approvals step rather than separate per-dApp approvals, reducing your standing on-chain exposure.

Filler and Solver Economics: Why Competition Matters

Both systems depend entirely on a competitive professional intermediary layer to pass value back to users. Without genuine competition, fillers and solvers would keep most of the available surplus.

UniswapX fillers earn the spread between the current Dutch auction price and their actual cost to source liquidity. On liquid pairs, that spread is thin — one to five basis points — so profitability requires winning many auctions and batching fills efficiently.

CoW solvers earn COW token rewards for winning batch auctions and capture any surplus above users' minimums, but the protocol's scoring function penalises over-extraction relative to competing solvers. The bonding requirement limits participation but also creates skin-in-the-game alignment.

Both ecosystems show real centralisation tendencies: a small number of sophisticated, well-capitalised firms win the majority of auctions. This is self-correcting in theory — a new entrant that undercuts incumbents will win volume — but only if entry barriers are low enough and order flow sufficient to justify the infrastructure investment.

Risks to Understand Before You Participate

Signature and approval risk. Signing an intent order grants the settlement contract spending authority up to the order parameters. A spoofed front-end could present a signed order for a different output amount or recipient than displayed. Verify EIP-712 structured data before signing. Use a hardware wallet with clear-signing support where possible.

Permit2 phishing. Permit2's SignatureTransfer mode is one-shot and non-replayable, but phishing sites have caused large losses by tricking users into signing permit messages for attacker-controlled addresses. In the first half of 2024, permit-style signature phishing on EVM chains affected around 260,000 victims for roughly $314 million in losses. Never sign permit requests on unofficial or cloned sites.

Order non-fill risk. If no filler or solver finds your order profitable — thin-liquidity pair, extreme volatility, or a trade too small to cover filler gas costs — the order expires unfilled. Your tokens are not at risk; the signed order simply lapses. But you may miss your execution window.

Filler/solver centralisation. Both protocols are operationally dominated by a small number of professional firms. Simultaneous withdrawal or collusion among that set would degrade execution quality or stop fills entirely.

Settlement-contract smart-contract risk. Both the UniswapX and CoW Protocol settlement contracts have been audited, but a critical bug in either would affect all orders routed through it. Audit history and sustained usage are indicators of robustness, not guarantees.

Off-chain relay integrity. Correct price discovery depends on the off-chain order relay being honest and available. A compromised relay could censor orders or selectively expose order flow to preferred fillers.

Unlimited Permit2 allowance. The initial Permit2 approval still requires an on-chain transaction and gas. Approving an unlimited allowance to Permit2 leaves ongoing exposure. Prefer a limited allowance where your wallet supports it.

Key Takeaways

• UniswapX uses per-order Dutch auctions with optional RFQ pre-fill. Competitive fillers settle early in the decay window, delivering price improvement above your stated floor. Best for large or time-sensitive single trades.
• CoW Swap batches orders and uses Uniform Clearing Price plus peer-to-peer CoW matching to eliminate intra-batch MEV structurally. CoW Hooks add pre/post-swap composability. Best for frequent traders and complex workflows.
• Both protocols redirect MEV surplus toward users rather than block builders, and both require a single approval step rather than per-dApp grants.
• Competition among fillers and solvers is the mechanism that delivers value to users — real centralisation in both ecosystems is the main structural risk.
• Signature phishing targeting Permit2 and EIP-712 flows is an active, well-documented threat. Verify every signing prompt on an official front-end.
• If you prefer a simpler entry point, you can swap inside Zelcore without interacting with these contracts directly. Part 3 of this series extends these single-chain mechanics to cross-chain execution via Chainflip and Relay.