I have an idea. It might be a crazy plan; I don’t know and I’m relying on you lovely people to let me know.
Full disclosure and transparency – I am not a coder or in any way advanced in terms of my knowledge about AI and its capabilities. That’s not to say I’m completely useless and have no idea what I’m doing, but let’s assume for now that what I think I know is actually incorrect.
I am in the process of starting a business which is a SaaS platform and as I have no money and getting funding for to build what is essentially an unvalidated product is proving difficult (classic chicken & egg scenario). So, I’d like to try and build the SaaS platform myself, or, to be more accurate I’d like to create a full stack development team comprising of different AI models.
But, let’s not stop there. Let’s recreate the entire real-world setting using AI models. So, what would that look like?
Well, you would have perhaps Claude given the role of backend engineer, maybe Gemini given the role of front end UX engineer etc. etc. Then, you need a set or peers to conduct peer reviews. All of the peer review roles could be given to Grok and his role would change depending on what he is peer reviewing. But, before all of that starts you need a group of subject matter experts from within the industry that the SaaS product is intended to be used, and these SME can be used during requirement analysis etc. etc.
If there was one central source of truth that contained all of the very detailed documentation that essentially is the blueprint of what is to be built, and all of the different AI models could access those documents, then they would all essentially be working from the same data set of knowledge from the perspective of what is to be built, what it must do, its functions etc. Then you could have a further AI model, possibly Copilot responsible for assigning each member of the team their daily tasks including prompt design, the team member would refresh its memory by accessing all the documentation, logs, etc. etc. and then carry out its task. At the end of the day what has been produced is peer reviewed which follows the same real-world process for peer reviews. The whole thing could to a large extent be autorotated as long as all the AI models are able to access one central point of truth from which they obtain all of their instructions and data etc.
So, am I mad? Does anyone know if this has been done before? It’s feels like a bit of an experiment. Is anyone up for joining me in this experimental adventure? As I say, I’m not a coder so I could really use the help 