[Advanced] — the attempt to fix wallet UX without giving up self-custody, and where it actually stands.

The problem it solves

Ethereum’s EOA has been unchanged since 2015, and it is a straitjacket:

  • One signature scheme (ECDSA on secp256k1). No passkeys, no biometrics, no multisig natively
  • You must hold ETH to do anything — including receiving your first token
  • One operation per transaction. Approve then swap is two signatures, two waits
  • No recovery. Lose the key, lose everything
  • No limits, no sessions, no policies. Every signature is unlimited authority

Every one of these is a self-inflicted wound from a design choice made once, and account abstraction is the effort to undo them without reintroducing a custodian.

The core idea

Make the account a smart contract, so its validation logic is programmable.

EOA:                       Smart account:
  fixed ECDSA check          YOUR code decides what a valid operation is
  ↓                          ↓
  "is this signature         "is this a passkey signature? is it under the daily
   from this key?"            limit? is it a whitelisted contract? did 2 of 3
                              guardians approve?"

Once validation is code, everything else follows: social recovery, spending limits, session keys, batching, gas sponsorship, any signature scheme.

ERC-4337 — without changing the protocol

Shipped March 2023, entirely in contracts and off-chain infrastructure — no hard fork, which is why it happened at all.

UserOperation           a pseudo-transaction: sender, calldata, signature, gas fields
      │
   alt mempool          a separate mempool, not Ethereum's
      │
   Bundler              collects UserOps, bundles them into ONE real transaction
      │
   EntryPoint           the singleton contract: validates each, then executes each
      │
   your Smart Account   validateUserOp() — YOUR rules
      │
   Paymaster (optional) sponsors gas — the user needs no ETH at all

The paymaster is the most immediately valuable piece. An app can pay its users’ gas, or accept USDC as payment for it. “You need ETH before you can do anything” — the single worst onboarding barrier — is simply removed.

Session keys are the second: authorise a temporary key with narrow permissions (“may call this game contract, up to this amount, for the next hour”). No wallet popup per action, which is what makes on-chain games playable.

The costs: a UserOp is more expensive than a plain transaction (the EntryPoint runs validation logic on-chain); bundler infrastructure is a new, partly centralised dependency; and it was a parallel system that ordinary EOAs couldn’t use.

EIP-7702 — the one that changed adoption

Shipped in Pectra (2025), and it’s the more consequential of the two for most users.

It lets an existing EOA temporarily set contract code, via a new transaction type. Your normal MetaMask address gains smart-account behaviour for a transaction, without migrating to a new address.

That last part is why it matters: ERC-4337 required moving to a new smart-account address, abandoning your history, your ENS, your NFTs, your positions. Nobody wanted to. 7702 removes the migration entirely — your existing address gets batching, sponsorship and session keys.

The two now compose: 7702 for existing EOAs, 4337’s infrastructure (bundlers, paymasters) underneath, with smart accounts still available for anyone wanting full programmability from the start.

What it enables, concretely

  • Sponsored gas — onboard a user who has never held crypto
  • Pay fees in USDC — no native token required
  • Batching — approve and swap in one signature, one confirmation
  • Social recovery — guardians (friends, devices, an institution) can rotate your key. The seed phrase stops being a single point of catastrophic failure
  • Spending limits and allowlists — a compromised session key costs you a bounded amount
  • Passkeys — sign with Face ID or a security key instead of a seed phrase
  • Session keys — playable games, tradeable interfaces, no popup per action

The honest status

What genuinely works today: gas sponsorship and batching are in production and materially improve onboarding. Safe has secured very large sums with smart-account logic for years. Passkey-based wallets exist and work.

What’s still rough:

  • Wallet and dapp support is uneven. Support for 7702 and 4337 varies by wallet, chain and library
  • Bundlers and paymasters are centralised infrastructure — another vendor in your “decentralised” stack → dapp architecture
  • Smart accounts are contracts, so they have contract bugs. A wallet with a vulnerability is a wallet that can be drained, and now your account has an attack surface. Use audited implementations (Safe, Kernel, Alchemy’s Light Account) — this is not the place to write your own
  • 7702 introduced its own risks. An EOA that can delegate to arbitrary code is a new phishing surface: sign the wrong delegation and your account executes hostile code. Wallets are still working out how to present this safely
  • Cross-chain address consistency requires care with CREATE2 and deployment ordering

For building

  • Use a provider — Alchemy Account Kit, Pimlico, Biconomy, ZeroDev, Privy. Running bundler and paymaster infrastructure yourself is a real operational burden with little upside
  • Sponsored gas is the highest-value feature to adopt first. It removes the biggest onboarding barrier for a modest cost, and it’s the easiest to add
  • Don’t force it. Support both EOAs and smart accounts; most users still arrive with MetaMask
  • Never write your own account contract

Key insight

Account abstraction turns “who may spend this” from a fixed protocol rule into application code — which is what finally allows recovery, limits, sessions and sponsorship without a custodian holding your keys. ERC-4337 proved it could work without a fork; EIP-7702 made it available to the accounts people already have, which is what actually moves adoption. The remaining barrier is that the infrastructure making it convenient is itself centralised.

↑ Back to Building Dapps