[idopmstuff]

I think this is a nice idea and decent proof of concept, but it's not good enough at anything to be useful right now. Everything it does can be done directly in ChatGPT/Code Interpreter.

There's definitely value to be created in being the layer that provides easy access to AI tools that are built to do the specific set of tasks an end user needs. On the other hand, it's going to be really challenging to compete with large incumbents who have both the distribution channel and the data. If Salesforce builds something like this, you don't have to give it all the company info that this tool is asking for, and anything it create can be immediately saved/used in the appropriate place for the business.

Integration with your existing data and services is really going to be key for this sort of thing - I would encourage anybody building something like this to have a clear strategy along those two dimensions before spending a lot of time writing code.

[joe_the_user]

"I think this is a nice idea and decent proof of concept, but it's not good enough at anything to be useful right now. Everything it does can be done directly in ChatGPT/Code Interpreter."

It may be that this may well wind-up being said about virtually every ChatGPT "app" that gets produced. After all, these apps essentially are just prepending some text to a ChatGPT prompt.

[redhale]

Prompts all the way down

[woliveirajr]

I just remembered this ancient comment on HackerNews :-)

"... you can already build such a system yourself quite trivially by getting ..." [0]

[0] https://news.ycombinator.com/item?id=9224

[idopmstuff]

Hah - I feel like mine is more reasonable than that, but human exceptionalism being what it is, I'm sure he would've thought the same. Still, I'd say asking ChatGPT to do these tasks is a fair bit more trivial than what that comment describes.

[joe_the_user]

The things that are specific about a "thin wrapper" to ChatGPt versus a "thin wrapper" to some other program or system, is that a ChatGPT wrapper has to end-up as "plain old English/language". This has the challenge that any/many people can generate this language 'cause language generate is what (nearly all) people do (while creating even a "trivial" command line string is a specialized skill) and also the challenge that such an app will inherit ChatGPT's quality of being never more than "mostly" reliable at any given task (as a result of GPT-3-4 being just a product of an amazing brutal large text averaging process).

[phillipcarter]

Yeah, it's a cool demo and proof of concept, and perhaps useful for casual users. Might even be a nice way to draw people into the platform for other things. But for most business use cases the integration with existing data and services is a lot (if not the bulk of the work), and can't be fulfilled by this.