a16z partner: Code is infinitely depreciating—why is it said that network effects are the ultimate moat in the AI era?

a16z Crypto CTO Eddy Lazzarin argued that as large language models drive the marginal cost of writing software toward zero, code itself is losing its value as a moat and the durable advantage is shifting to "reusable networks" — deep liquidity, immutable reputation, portable identity graphs and settlement credibility that cannot be cloned by AI. He pointed to Uniswap as an early example of composability, where new apps could plug into existing liquidity rather than rebuild pools, but noted that onchain experimentation was previously too costly for all but a small group of developers. AI has collapsed that barrier, letting even non-technical builders ship working prototypes in hours and pushing a wave of experimentation across NFTs, DAOs, decentralized social networks and fully onchain games. Lazzarin's core claim is that AI can regenerate any front end or API in minutes but cannot conjure the provenance, community consensus, onchain identity, credit reputation or pooled liquidity behind a network — "you can't vibe-code a USDC." He frames blockchain as the internet's canonical shared-state layer, arguing that as applications become commoditized, value capture migrates to the networks they compose on, and that this shared state becomes more essential in a world of AI agents transacting on behalf of users. The takeaway for builders is to target networks with strong network effects rather than interfaces, though the piece offers no metrics on adoption or usage to test the thesis.