Following up on a prior post - I've been reflecting on what makes an LLM-based product or service defensible. My current conclusion is that if you're building a GPT wrapper you should lean into system of records or integrations into the relevant ecosystem so you're not as easily replaced by GPT-5 or a new OpenAI feature.

Any app that's just text in text out, at the limit, can be replaced by ChatGPT. But if you're maintaining state, tracking things over time, integrating deeply into the user's other systems, and embedding into the user's actual workflow you have a much better chance of long term success.

The flip side to that is when integrations and system of records are the distinguishing feature, the incumbents have the advantage.

Is it easier for them to integrate AI or for you to build their integrations and distribution?

There's no right answer here, but I think that means you have to choose your market and competitors wisely. Things that might work, choosing:

• competitors that will have a hard time integrating AI either due to culture or technology
• non-winner-take-all markets
• greenfield markets where the product wasn't possible pre-GPT