← Research · Strategy
Defensibility in AI Products
A durable product gives customers a reason to stay that survives a model or feature comparison. Test whether workflow fit, trust, useful data, or distribution improve with real use.
By Jake Bauman · Revised
Jake Bauman's strategic checklist, drawing on product and agent design sources listed below.
Separate a feature from an advantage
List what a competitor could reproduce with the same model. Then list the customer knowledge, workflow design, permissions, and service quality that require sustained work.
Look for an improvement loop
- Can users correct an output?
- Does the product remember the correction with consent?
- Does the next result improve in a way users notice?
- Can you measure retention and trust without treating usage alone as success?
Limit of this framework
A feedback loop is not automatically a network effect. Some corrections help one account only, and privacy constraints may prevent shared learning. Test the mechanism before claiming a moat.
A claim to test
If you believe your product improves through use, name the precise feedback signal and where it changes the next result. A saved customer preference can improve one account without creating a broader advantage. Ask whether customers notice and value the improvement. Then test whether they continue using the workflow when a competing model becomes available.
Sources and context
These sources inform the framework. They do not validate the suggested actions for every business.
- Brian Balfour, Four Fits for $100M+ Growth: Original source of the Four Fits framework.
- Anthropic, Building effective agents: Provider guidance on agent design and evaluation.
Want a practical starting point? Enter your email to get the free Growth Agent Starter Kit download link on the page.