A developer who bet Vitalik Buterin that a single person could agentic-code an Ethereum (ETH) client targeting the 2030 roadmap built ETH2030 in two weeks - 702,000 lines of Go, covering 65 roadmap items, syncing with Ethereum mainnet.
Buterin commented on the result Friday, calling it "quite an impressive experiment" while listing significant caveats, and raising the possibility that AI could accelerate the Ethereum roadmap's completion more than the community currently expects.
Buterin was direct about what the project is not: built without formal Ethereum Improvement Proposals in place, it almost certainly contains critical bugs and likely includes "stub" versions of features where the AI did not attempt a full implementation.
The point, he argued, is the trajectory rather than the output. Six months ago, he wrote, even a prototype of this scope was "far outside the realm of possibility."
Speed vs. Security
Buterin framed the right approach to AI-accelerated development as splitting the gains between speed and security rather than spending them all on velocity.
His preferred model: use AI to generate larger test-case suites, formally verify implementations, and produce multiple independent versions of the same component for cross-checking.
A collaborator on the LeanEthereum project - which aims to formally verify all Ethereum code - recently used AI to produce a machine-verifiable proof of one of the most complex theorems underlying STARK security.
That, in Buterin's framing, is the higher-value use case: not shipping faster, but shipping more verifiably correct code.
Read also: Kalshi Refused To Let Traders Profit From Khamenei's Death - Then Refunded Everyone On A $36M Market
What He Thinks Is Possible
Buterin was careful to frame his optimism as a possibility rather than a forecast. He said people should be open to the Ethereum roadmap finishing faster than current timelines suggest - and at a higher security standard.
On the security front, he expressed personal excitement about the potential for bug-free code to transition from "idealistic delusion" to a basic expectation for critical infrastructure.
He acknowledged that total security remains impossible in any absolute sense - it would require exact correspondence between code and the contents of a developer's mind - but argued that AI formal verification can already eliminate more than 99% of the negative consequences that could follow from specific, well-defined code failures.
The ETH2030 repository remains publicly accessible at github.com/jiayaoqijia/eth2030.