Moats in the Age of AI

• Switching costs—historically THE moat for SaaS—are getting demolished as AI agents automate migrations, rewrite integrations, and compress what took months of consultants into weeks
• Scale and brand both split: application-layer scale advantages weaken while infrastructure-layer strengthens; marketing-driven brand compresses while institutional trust brand becomes more valuable in high-stakes contexts
• Network effects split too: shallow exclusivity crumbles (agents can multi-home and arbitrage marketplaces) but deep liquidity networks with trust, reputation, and coordination density remain durable
• Cornered resources (especially proprietary data) become MORE valuable—but AI exposes fake moats where data was actually scrapeable all along
• Counter-positioning remains powerful for startups: offer usage-based pricing vs per-seat, replace workflows with agents vs augment users—incumbents struggle to adopt without cannibalizing revenue

The prevailing narrative says AI kills all software moats, but a systematic walk through Hamilton Helmer's 7 Powers framework reveals a more nuanced reality: labor-based advantages collapse while structural advantages persist or strengthen. The biggest casualty is switching costs—the default SaaS moat—which AI agents demolish by automating schema mapping, rewriting integrations, generating training materials, and running parallel systems. What once required months of expensive consultants compresses to weeks of automated orchestration. Meanwhile, if agents handle workflows, human-centric UI/UX becomes less relevant as a lock-in mechanism.

Scale economies and brand both bifurcate. Application-layer scale advantages (spreading R&D across users) weaken as 20-person teams with agents match the velocity of much larger orgs. But infrastructure-layer scale (OpenAI, Anthropic, AWS) strengthens. Similarly, marketing-driven brand weakens as agents can systematically evaluate price/performance tradeoffs, but institutional trust brand becomes more valuable in high-stakes contexts with AI unpredictability and hallucination risks. Network effects follow the same pattern: agents enable frictionless multi-homing and can arbitrage across marketplaces (imagine voice agents calling restaurants not on your platform), but they can't fabricate real-time liquidity, courier density, or canonical reputation graphs—marketplace density and trust remain structural.

The moats that strengthen are cornered resources (proprietary data becomes MORE valuable for training models, but AI exposes datasets that were never truly proprietary), counter-positioning (startups can offer usage-based pricing and agent-first workflows that incumbents can't adopt without cannibalizing revenue), and process power built on proprietary data or institutional knowledge. The playbook for AI-native companies: you can't rely on switching costs anymore. Build around proprietary data, real network effects, and compounding process advantages—being extremely AI-native from day one is the new table stakes for process power.