What is defensibility?

Defensibility comes from the competitive advantage that a company or product has over its competitors. These advantages operate like “moats”, protecting a company against its competitors. Trusted brand or switching costs are good examples of moats that competitors find hard to copy or emulate.

How should PMs think about defensibility?

Being unique in the market or distinctly better at solving a problem is only meaningful if you can successfully defend that position. To achieve this, defensibility should be designed into the product from the start, not bolted on later. There are lots of different areas where your product could have a defensible advantage, but let me share some of the most common ones:

• Network effects – Product features where each additional user makes the product more valuable. Think of examples like LinkedIn, Uber, Slack and WhatsApp that are designed to seamlessly add and connect new users so that their networks can grow faster than the competition.

• Switching costs – Think of products like Salesforce and Shopify that are deeply embedded in sales and marketing workflows. As a result, companies and users become locked into these products and their systems.

• Proprietary data – High quality, proprietary data can be generated through product usage (e.g. feedback, behavioural traces, domain‑specific content) that in turn improves the product experience over time. Duolingo, TikTok, Granola and GitHub Copilot are good examples here.

• Brand reputation – Think of brands like Microsoft 365 and Apple that have built a strong brand over time, with a track record of strong performance and product experience respectively, which gives them a distinct competitive advantage that is hard to emulate.

I often link moats to clarity about the competitive field where your product plays (and where it has a chance to win). Products have the best chance of competing successfully in a market where their solution gives them an ‘unfair’ advantage; providing a unique or significantly better solution for a specific customer pain point.

How to build defensible AI products?

I’ve written before about the risk of companies simply creating ‘AI wrappers’; applications that are primarily built on frontier models and their APIs. There are three main moats that AI products can build in order to become defensible:

• Infrastructure ownership – Infrastructure companies win because others build on them and become dependent on them. Think AWS, Snowflake, or Stripe—the invisible layers that only matter when they break. Ask yourself: Would anyone rebuild my product if it disappeared? Would anything break if it went away? If the answer is no, your product isn’t infrastructure, and it’s unlikely to become durable.

• Embedded in workflows – When AI products are successfully embedded in users’ workflows, users rely on the product to achieve key outcomes. Products like ComplyAdvantage, Taktile and Harvey are designed to be part of critical, daily workflows. Companies like Salesforce, Ramp and OpenAI are increasingly using Forward Deployed Engineers, software engineers who work hand in hand with customers to map bespoke workflows and automate them quickly.

• Unique, proprietary data – We are seeing the powerful things that AI frontier models can do with data, but that doesn’t necessarily give companies a moat. Access to specific data, be it private customer or industry specific data, will give companies a competitive edge in the AI era. This data is often unstructured or hard to access, hence why buying or integrating specific data sets is becoming increasingly important, using the data to train your LLMs to deliver specific, quality outputs. This data then gets compounded through user feedback, annotations and corrections generated by users.

Main learning point: Building defensible products isn’t about being first to market or having the most features—it’s about designing moats into your product from day one. Whether through network effects, switching costs, proprietary data, or brand reputation, PMs need to answer one critical question: “What would make it painful for customers to switch to a competitor?” In the AI era, this defensibility increasingly comes from owning infrastructure, embedding deeply into workflows, or building unique data advantages that compound over time.

Related links for further learning:

1. Desirability vs Defensibility by Jason Knight
2. Revenge of the GPT Wrappers: Defensibility in a world of commoditised AI models by Andrew Chen
3. Building a moat in the age of AI by Matt Gatto and Jessie Sheff
4. Playing to Win: How Strategy Really Works by A.G. Lafley and Roger Martin
5. How to build defensible products when shipping is cheap by Frankie Cleary
6. What are Forward Deployed Engineers, and why are they so in demand? by Gergely Orosz
7. Data Moats in Generative AI by Kenn So
8. Understanding the Forward Deployed Engineering (FDE) Model by Maxim Atanassov
9. Forward Deployed AI Engineer by Sundeep Teki
10. OpenAI Product Leader: The 7-Step Playbook for Defensible AI Products by Miqdad Jaffer and Peter Yang