To avoid being made obsolete by a frontier AI model, startups need a strong moat. The three most defensible moats are: 1) building hardware, which AI cannot physically replicate, 2) establishing strong network effects where value increases with more users, and 3) operating in a complex, regulated industry requiring human interaction.

While AI agents could shift sales away from traditional retailers, companies with extensive physical infrastructure and forward-positioned inventory have a defense. AI agents prioritizing speed and efficiency for physical goods will likely still favor these established networks, preventing full disintermediation in the new agentic commerce landscape.

In an era dominated by AI, businesses requiring physical infrastructure and specialized, licensed human intervention (like doctors or pharmacists) are highly defensible. AI can expand the top of the marketing funnel, but the company controlling the real-world delivery and expert services captures the value.

Marketplaces like DoorDash are more than just software; they are logistics and customer service networks that solve messy, real-world problems. An AI agent can discover a restaurant, but it cannot handle a cold sandwich or a refund, giving these physically-integrated companies a durable moat against pure software disruption.

The term "unsloppable" describes companies whose competitive advantage isn't their codebase, which AI can replicate. Instead, their strength comes from durable moats like hardware, strong network effects (Uber), exclusive IP (Disney), or physical infrastructure, which are difficult for AI-powered startups to clone.

As AI commoditizes software, the most defensible businesses are no longer asset-light SaaS models. Instead, companies with physical world operations, regulatory moats, and liability are safer investments. Their operational complexity, once a weakness, now serves as a formidable barrier against pure AI-driven disruption.

Investor Henry Ellenbogen favors two types of competitive advantages. First, hard-to-replicate physical assets like distribution networks, which are messy and time-consuming to build. Second, “soft” moats built on elite human systems for talent development, operational excellence (like the Danaher Business System), and sharp capital allocation. These are harder to see but just as powerful as physical scale.

To avoid being disintermediated by AI agents that could direct consumers elsewhere, retailers can leverage their physical assets. An AI agent will still prioritize retailers with extensive infrastructure and forward-positioned inventory to ensure fast and efficient delivery, creating a competitive moat against pure-play e-commerce.

Oren Zeev argues against the narrative that AI will kill all incumbents. He believes businesses with operational complexity, deep data moats, and strong distribution are not easily disrupted. These companies are more likely to leverage AI to their advantage, while simpler software companies are at greater risk.