At least 14% of top Crypto open source projects have used AI to review or generate code AI assisted programming is rapidly becoming no longer just a gimmick, but deeply integrated into the development practices of the encryption industry. After analyzing the 1000 most followed (with the highest number of stars) encryption related code repositories on GitHub, I discovered an important trend: >At present, at least 137 code repositories (about 14%) clearly show signs of using AI assisted programming tools. AI tools have been integrated into various key tasks of encryption projects, including automated code review, bug and vulnerability detection, code assistance and auto completion, and multilingual component translation. Among these AI adopters, Claude and GitHub Copilot dominate the market, accounting for approximately 87% of AI usage in total. OpenAI's Codex, cursor, and Google Gemini, among other participants, have a relatively small share. Given that these findings rely solely on clear and verifiable AI usage metrics (such as mentions in submitted information or pull requests), this 14% adoption rate is likely only a conservative lower bound estimate. Considering that many developers use AI tools locally or through IDE integration, and these behaviors do not leave traceable traces in the code repository, the actual AI usage rate may be higher. AI is no longer limited to peripheral scripts, but has entered core protocols and client implementations, and is widely used in wallets, security tools, and even ecosystem documents. Layer 1 and Layer 2 protocols: - ethereum/go-ethereum - aptos-labs/aptos-core - MystenLabs/sui - near/nearcore - stellar/stellar-core - ethereum-optimism/optimism Wallet: - MetaMask/metamask-extension - rainbow-me/rainbow - WalletConnect/walletconnect-monorepo - BlueWallet/BlueWallet Security and Audit: - trailofbits/algo - crytic/solc-select Documents and specifications: - ethereum/EIPs - ethereum/ethereum-org-website - MetaMask/metamask-docs - duneanalytics/spellbook - ethereum/execution-specs - foundry-rs/book Organizations often choose multiple AI tools and carefully match them based on the technology stack, platform, and use cases of each project. for example MetaMask: -Metamask extension (using cursor) -Metamask docs (using cursor) -Metamask mobile (using Claude) a16z: -Helios (using Claude) -Halmos (using Copilot) -Jolt (using OpenAI) Coinbase: -X402 (using cursor) -Onchainkit (using Copilot) The methods used in this study include: one ️⃣ Collect GitHub code repository data related to encryption from over 500000 records provided by Electric Capital (Crypto Ecosystems). I first collected the number of Stars for these code repositories and filtered out the top 1000 based on their popularity. Code repositories that have been archived (indicating that they will no longer be updated) are excluded. two ️⃣ Analyze the latest 200 commits and 200 merged pull requests in each selected main branch of the code repository, and search for keywords related to AI programming agents such as Claude, Copilot, or Codex. three ️⃣ Perform manual verification after initial keyword filtering to avoid false positives. For example, the keyword "gem" may refer to a cryptocurrency exchange or Google's AI model. Despite taking these preventive measures, achieving absolute accuracy remains challenging. As this is an open research project, further clues or suggestions are highly welcome. The rapid popularization of AI tools may reshape the competitive landscape of encryption development. Effectively utilizing AI in projects can accelerate development speed, improve code quality, and quickly adapt to emerging security threats. In short, AI assisted programming is the trend, and its influence will only continue to grow. For cryptocurrency projects and investors, maintaining a leading position in this field may soon no longer be just icing on the cake, but rather crucial. Data source https://substack.chainfeeds.xyz/p/14-crypto-ai