To make the most effective use of AI, you pretty much have to be an expert in the domain you are working so you can write effective prompts AND AI must have enough relevant information in order to provide quality output. There is a lot of info in telecom that cannot reasonably be put into AI (confidential info, or spread across many emails, and across many documents (each containing PPI), etc.). For example, take a cell site design: To use AI you would need to upload many plots from Atoll, upload traffic usage and drive data (if available), upload maps, upload ongoing surrounding project info and their status', etc.. You would also need to tell AI what you want as far as standard equipment (and that is always changing and has dependencies), and on and on it goes. The list of things AI needs in order to provide quality output goes on and on, depending on the domain and the intricacies of the required output. The end user also must be knowledgeable enough to recognize garbage.
If you are a non-tech type (not intending an insult here), and you took some of the AI courses available from Google (and probably other AI vendors), you have seen some pretty powerful stuff. AI can do some amazing things. Yes, AI can summarize data. AI can create graphics, videos and audio output. AI can tie together data and create powerful dashboards. AI can write code. AI does all of this much faster than humans. Those AI courses leave the user thinking AI can do most anything. Those AI courses also inspire users to want to buy the creator's product (AI subscriptions). That is intentional! Google, and other AI vendors want to make money on their products. They are not going to talk much about the pitfalls of their products.
Taking those courses is like watching a very long, well produced commercial. If you are a tech type, you may be familiar with at least some of the inner workings of AI and understand it really is a probability machine (really A LOT of little probability machines). It takes in data, creates relationships amongst that data based on probabilities it "learned" from training data. Some of the magic is taken way, but it is a good thing to understand in order to make better use of the tool.
Dan Schulman's educational background is in economics and he has an MBA. Guessing other C-suiters' educational backgrounds are similar. AI works well for economics. You can upload many disparate spreadsheets and as long as there is some semblance that the data in each sheet can be correlated with data in the other sheets, AI will handle it like the Harlem Globetrotters handle a basketball. I am sure Dan was drooling at the mouth when AI vendors showed him what they could do for VZ. But, give AI a bunch of data without concrete direction, and design requests that could have many tradeoffs that are not seen until design time, it will provide output that is not concrete. There will be a lot of garbage output.
To best use AI, yes, you can use conversational prompts, but, you must be very precise. Think of training your dog or you child:). Dogs and children do not understand the concept of "sometimes." You must think like a programmer and tell AI exactly what you want. The issue with more complex problems is that exactly what we want is not known until we get deeper into the design. We have not seen all of the dependencies and tradeoffs until we get deeper into the analysis. By the time we get that far into it, it is faster to just do the task ourselves rather than try to put all of that info into AI.