Writing a lending protocol only takes an afternoon: In the AI era, which three moats does DeFi rely on to actually make money?

AI coding tools like Claude Code have collapsed the cost of building DeFi protocols, with founders now able to ship a lending engine or perpetuals exchange in a weekend versus the $3 million and six months required five years ago, but Multicoin's Spencer Applebaum argues this makes distribution, capital, and proprietary data the real moats. Distribution is already proving decisive: Hyperliquid holds over $6 billion in USDC as derivatives margin and, under its AQA framework deal with Coinbase, now captures roughly 90% of reserve yield on about $5.5 billion of that balance—around $185 million a year flowing to its Assistance Fund for HYPE buybacks—rather than to Circle. Wallets out-earn venues, with MetaMask charging up to 0.875% on swaps versus Uniswap's 0.13% average fee over the past 12 months, most of which goes to liquidity providers, and Phantom adding 5 basis points on top of Hyperliquid's native 4.5 bps perps fee. The backdrop is DEX spot share of CEX volume hitting a record 24%, alongside growing tokenized RWA adoption by traditional institutions. The takeaway is that value is shifting from open-source protocol code to whoever controls order flow and balances, so the next thing to watch is whether wallets and frontends internalize order matching and whether large balance holders like Hyperliquid keep extracting yield concessions from stablecoin issuers.