ERC-4337 versus native account abstraction on zkSync and Starknet
Account abstraction is not one thing. It is an idea with two serious implementations, one living at the application layer and the other living in the chain itself. They solve the same problem - letting wallets behave like smart contracts rather than simple keypairs - but arrive at very different trade-offs.
What ERC-4337 actually is
ERC-4337 is an Ethereum standard. It works without changing the Ethereum protocol. No hard fork. No new opcode. It exists entirely as a set of smart contracts and off-chain infrastructure.
The core mechanism is the UserOperation. Wallets submit these objects to a global mempool. Bundlers scoop them up, validate them against an EntryPoint contract, and package them into regular Ethereum transactions. Paymasters can sponsor gas. Users can pay in ERC-20 tokens. Signing logic can be anything the wallet contract allows.
The strength is portability. Because nothing changes at the protocol level, ERC-4337 works on any EVM chain - Ethereum mainnet, Polygon, Arbitrum, Optimism, all of them. A wallet built for ERC-4337 can live across dozens of networks without modification.
The weakness is the bundler. ERC-4337 adds a new middleware actor. Bundlers must be operated by someone. They must be trusted to include UserOperations fairly, and they introduce latency and cost that do not exist when the chain itself handles account logic.
Native account abstraction: zkSync and Starknet
zkSync and Starknet took a different route. They built account abstraction into the chain's core. Every account on these networks is a smart contract by default; there is no separate concept of an externally owned account, no EOA at all.
This changes everything. The chain's sequencer or validator directly processes account logic. No bundler. No separate mempool for UserOperations. The verification, execution and fee payment all happen at the protocol level in a single atomic step.
The efficiency gain is real. Native AA avoids the overhead of packing and unpacking UserOperations, sidesteps bundler competition and potential censorship, and lets wallets implement recovery, session keys or multi-factor signing without relying on off-chain infrastructure to make it work.
The cost is chain-specificity. zkSync and Starknet are themselves EVM-compatible in limited ways. A wallet written for zkSync's native AA does not work on Ethereum mainnet. The approach is not portable. It ties developers and users to a particular Layer 2.
Concrete examples: Argent and Braavos
Argent began on zkSync as a native AA wallet and later deployed to Starknet using that network's own native AA system. Braavos is a Starknet-native wallet that relies entirely on the chain's built-in account abstraction features.
Both wallets offer features that ERC-4337 wallets can match only through bundlers and external contracts. Argent uses native account logic for social recovery and daily spending limits. Braavos uses Starknet's native fee abstraction to let users pay gas in tokens other than ETH. Neither needs a paymaster contract or a bundler to do it.
The trade-off is clear. If you want those features on Ethereum mainnet, you use ERC-4337 and accept the bundler dependency. If you want them with lower overhead and tighter integration, you pick a Layer 2 that supports native AA - and accept that you cannot leave it.
Which one wins?
Neither. Or both. It depends on what the user or developer needs.
ERC-4337 is the portable standard that works everywhere Ethereum runs. It is immature: the bundler market is small, the mempool is not yet fully decentralised, and real-world throughput is low. But its reach is unmatched.
Native AA on zkSync and Starknet is faster and cleaner, doing what account abstraction was supposed to do without adding new infrastructure layers. But it only works where it was built. A Braavos wallet does not move to Arbitrum.
The market is still early. Most users do not interact with account abstraction at all. Those who do are choosing between a general-purpose standard with middleware costs and a chain-specific standard with none. It is a split that will probably not last. Eventually, Ethereum may integrate account abstraction at the protocol level itself. Until then, you pick your trade-off.
Not financial advice. meow-cto.xyz publishes market data and general information about Meow. Crypto assets are volatile and you can lose everything you put in. Nothing here is a recommendation to buy, sell or hold, and we make no price predictions.
Prices are sourced from third parties and may be delayed or wrong. Verify anything you intend to act on against a primary source.