A common reaction to the ability of AI models to write software is “I can create apps, sell them as products!”.

On second consideration it should be clear that if you can build an app that is valuable to someone just by prompting a model, so can everyone else.

To create something with unique value, some kind of edge is required. It could be niche domain knowledge, a proprietary dataset, an advanced technical component that the model is not (yet) capable of achieving.

Alternatively, when the model causes the cost of software production to drop dramatically, we can stop thinking about the app that brings value to many, and think about the app that brings value just to ourselves- personal software.

This talk discusses some of the advantages of building software for one’s own use, how it avoids some of the complexities of products:

- Single platform (“I have an iPhone”)
- No requirement for monetisation (ads, subscription integration not required)
- No need for scale on the server side
- No need for accessibility and internationalisation beyond your own need

And brings some additional advantages
- No user tracking
- No malware, spyware, etc risk
- No incentive to focus on “user engagement”, addictive dark patterns and so on.

The talk then proceeds to demonstrate an example, using a proposed architecture suitable for personal software

- Progressive Web App deployment (no need for App Store integration)
- Standard web technologies (play to AI Model’s strengths)
- Serverless deployment pattern (unlikely to exceed cloud platform’s free tier on a personal app)

Finally the talk explores the implications for software-as-a-service platforms. Which are safe? Which are at risk? How much better do models need go get, to threaten a typical SaaS business?