The phone company always wins

• Agents commoditize aggregation (showing all options) but can't replicate true network effects (where value comes from multiple parties transacting)
• DoorDash's moat isn't listing restaurants—it's coordinating drivers, bundling orders, and managing reputation across a network
• Agents expand network effects in three ways: becoming network nodes themselves, lowering onboarding friction to near-zero, and using cheap aggregation to bootstrap multiplayer networks
• The cold start problem gets solved: start with useful single-player utility, let the network form as a byproduct (like DoorDash manually calling restaurants before they signed up)
• Software margins are compressing, but network effects remain the path to durable businesses—just with different GTM (UI-less onboarding, agent-to-agent transactions)

The conventional wisdom that "agents destroy moats" misses a crucial distinction: agents destroy aggregation moats, not network effects. Aggregation—putting all options in one place—is what agents commoditize. A vibe-coded delivery app can surface restaurants trivially. But true network effects, where the product improves because multiple parties use it simultaneously, not only survive but become viable in more categories. DoorDash's real value isn't showing you restaurants; it's coordinating drivers for better routing, bundling orders for efficiency, and managing reputation across transactions. These multiplayer dynamics get stronger with density, and agents can't replicate them.

Agents actually expand the surface area for network effects in three specific ways. First, agents themselves become nodes in networks—agent-to-agent transactions create coordination layers that compound faster than human networks because machines can participate at volumes and speeds humans never could. Second, agents lower onboarding costs to near-zero, activating dormant network effects in markets previously bottlenecked by UI complexity. People can interact with networks invisibly through agents instead of wrestling with interfaces. Third, cheap aggregation becomes a bootstrapping mechanism: you can start with genuinely useful single-player utility and backflip into a multiplayer network as a byproduct, solving the cold start problem that killed marketplaces for decades (like DoorDash manually calling in orders before restaurants signed up).

The economic constraint on AI displacement provides a natural boundary: if automation expands rapidly, compute demand rises, pushing marginal costs above human labor for certain tasks. This means software companies will look different—not pure application software or traditional SaaS, but businesses with permanently lower margins that still throw off huge cash by capturing value through network effects. The playbook shifts from defending aggregation moats to using agents as an on-ramp for building real networks, with new GTM approaches like UI-less participation and single-player-to-multiplayer flips.