Y Combinator on moats: when it matters, when it misleads

This article summarizes an episode of Y Combinator’s video series featuring its partners, Gary Tan, Harj Taggar, Jared Friedman, and Diana Hu.

Y Combinator CEO and President Garry Tan / Photo credit: Tech in Asia

Obsession with “moats” has become a common reason many founders fail to even get started. In a recent discussion, Y Combinator partners Gary Tan, Harj Taggar, Jared Friedman, and Diana Hu explored this founder anxiety, offering a counterintuitive perspective on when and how to think about building a competitive advantage.

Building a moat around nothing

Founders often get sidetracked by building defenses before they have a product. The group agreed that the only initial priority is to find a customer’s pain point and build a solution as fast as possible.

First, build the castle
Taggar explains, “a moat is inherently a defensive thing, and you have to have something to defend… if you got nothing to defend, don’t worry about your moat.”

Speed is the only early moat
Diana Hu argues that before finding a product people want, a startup’s only real advantage is its pace.

“At the beginning, the only [moat] that startups have is really just speed. Once you pass that and build something that people want, then you figure out and go deeper into these types of moats.”

Premature optimization is a trap
Worrying about how to protect the business in the future can become an excuse for not starting at all. Friedman warns against this inaction from overthinking.

“They try to use [moat analysis] to pick between two different startup ideas because they’re trying to forecast five years in the future which one will have a greater moat, which just isn’t how it works.”

A modern take on a classic playbook for AI

Once a startup has something valuable to protect, the group points to Hamilton Helmer’s book, Seven Powers, as an essential guide. While its examples are from the pre-AI era, Friedman notes that the core ideas are timeless.

• Process power: Building a system so complex it’s hard to copy. For an AI company, this means an agent with logic developed over years to handle real-world exceptions reliably.
• Cornered resources: Securing special access to a valuable asset, like customer data from workflows and user activity.
• Switching costs: Making it too difficult for customers to leave. For an AI company, this is achieved through integrations or personalization that makes the agent invaluable.
• Network economies (network effects): The value of a product increases as more people use it. In AI, this is often driven by data. More usage generates more data, which improves the AI model for all users.
• Scale economies: A business with a large scale of operations can offer a service at a lower cost per unit. For AI, this is most evident at the foundation model layer, where the massive upfront cost of training allows large labs to run their models at a price new companies can’t match.
• Counter-positioning: Adopting a business model that established companies cannot copy without damaging their own business. For AI startups, this often means pricing based on value or tasks completed, directly challenging the “per-seat” pricing of established SaaS companies.
• Branding: Building a strong reputation that makes customers choose your product over an equivalent one, even when competitors have similar technology.

The moat is built on boring work
While an AI demo can be built in a weekend, Friedman and Taggar point out that the work lies in making it reliable for customers. The moat isn’t the initial idea, but the difficult final stretch.

Friedman warns, “The version you build in a hackathon isn’t useful to anyone.” To which Taggar adds, “The last $10\%$ of getting [an AI tool] to work reliably across tens of thousands of KYC requests per day is a particular type of painstaking drudgery work, in a way that lots of engineers are just not excited to do.”

AI puts big companies in a corner

Building a great product is one way to win. Attacking a flawed business model is another. AI startups are positioned to do both, as many large SaaS companies are stuck with pricing models that AI is set to make obsolete.

A crisis of engineering culture