The point is that it simplifies the codeing. If you do this via older methods the coding gets very complicated quickly and have difficulty handeling input that isnt very specific. The point about AI systems is that they can take fairly vague input like human beings asking questions without all the specifics and generate an answer. Without it you have to be fairly specific and consistant in what you ask and in exactly the right format. Think the voice command systems in cars the older ones require you to ask things in exactly the format that was written by the programers.
I'm not an AI conspiracy theorist, I work with this stuff all day every day and specifically with speech.
You're right that it can do a better job figuring out what you said but it doesn't simplify coding. I'd make an argument that by the time you get the guardrails in place it's quite a bit more complex.
The older methods aren't any more complicated than using AI and those tools still continue to get improved. You get more precise control ... without all the expensive inference overhead.
If you're a computer guy, checkout "vosk", it's a very acurate and incredibly light weight speech recognition system. Inference requires a lot of compute and to no real advantage here, a solid vosk implementation would get you there with amazing acuracy.
The actual problem with speech recognition, especially in our small airplanes and cars, in my humble experience (been experimenting since probably 1982, professionally since 2005) is that the same basic problem exists ... it works wonderfully in perfect conditions ... but once you're on bluetooth speakerphone in your car with kids screaming in the back seat you're lucky to get 20% accuracy. Imaging how great it's going to work with our small airplane noise.
Don't forget, these models don't want to tell you "I can't" or "I don't know", they will lie.
While incredibly useful in the right context, I don't think there needs to be a mad rush to get AI into everything.