"Be Different" doesn't work for building products anymore (iamcharliegraham.substack.com)

AI-assisted coding (examples: Claude and “vibe”-based coding assistants) has dramatically shortened the product development lifecycle—what once took months can now be sketched, coded and deployed in days—creating a Cambrian explosion of lookalike software. That acceleration (developers building ~5x faster) plus easy reproduction of UX (screenshots or soon video→AI replicas), synthetic or similar-enough data (~80% utility), and cheap wrappers means features, flows and business models are copied in days or weeks. The result is a massive red ocean where traditional differentiation—better UX, single standout features, or “proprietary” data—no longer buys long-term advantage. Expect rapid technical debt from quick builds and a “vibe wall” for non-developers, but overall competition intensifies and lone voices get drowned out. In this environment, sustainable moats come from distribution and structural complexity rather than product novelty. Defensible strategies include massive existing distribution channels, obscure/regulated niches that require deep domain knowledge, products with expensive or bespoke integrations, true large network effects, compounding data lock-in (system-of-record friction), and regulatory barriers that raise entry costs. Big platforms will also absorb many standalone tools by bundling “good enough” AI features. For AI/ML builders and founders the takeaway is tactical: pick niches with high integration or regulatory friction, build real network effects or lock-in, or leverage pre-existing distribution—otherwise you’re competing in a race to be copied.