The author expresses skepticism about the idea that large language models (LLMs) will displace software-as-a-service (SaaS) products. They argue that while LLMs can reduce the cost and time of building software, many SaaS offerings provide essential ongoing value like compliance and updates, which LLMs can't easily replicate. However, after experiencing frustrations with a small SaaS called Shoutout.io—especially its broken billing feature—the author decided to replace it using LLM-generated code. This process took only 20 minutes. The author outlines their steps to recreate the testimonials feature of Shoutout.io using Codex, an LLM tool. They created a simple setup that stored testimonials in a JSON file and generated HTML for display. The outcome was visually identical to the original service, but without the dependency on a third-party provider. While the author acknowledges that rebuilding complex SaaS products remains challenging, they highlight the risks for those that don't offer ongoing value. Shoutout.io's static functionality made it easy to replace when issues arose, emphasizing that customers may increasingly look for alternatives when faced with poor service. The piece suggests that the landscape for SaaS could change as customers become more willing to switch to DIY solutions. The author notes that issues like broken features can drive users away, especially if they can create a simpler version quickly with tools like LLMs. The article raises questions about the long-term viability of some SaaS businesses, particularly if they don't invest in maintaining product quality and customer satisfaction.