I already see that. It is gonna be a massacre.
Of course it won’t. Even in the video you link at 20:49: “A lot of people are worried that it’s gonna take over my job and then I’m not gonna have the work, but you’re still are involved in the process. It’s just helping you do your job better and faster”.
Which is painfully obvious even with al the amazing generative tools that are shown in the video.
Because there are a hundred million small and large decisions that have to be made that the AI cannot do until the person who knowns about these details tells it.
— long story —
I don’t know that much about construction, so I’ll give you an example from my area. I work for Spotify. Let’s say you’re tasked with implementing a “skip to next song” button. A surprising amount of people today will say that “yup, your job is gone, the AI can do that”. Whereas a skip button will depend:
- on auth state
- type of account (premium/ad-supported)
due to licensing on some devices (only some devices, and often not even all devices from the same brand) you cannot skip more than a certain number of songs, and that should result in a valid message to the user (for example, in a voice saying “you cannot” on smart speakers)
- current player states (end of queue, end of playlist, depending on quite a few options within the player and the device)
- media availability due to licensing, country, external and internal providers. You need fallbacks for that.
- partner integrations (e.g. running on various brands of smart TVs will have their own slightly different issues with UI, API calls, displaying data on screen etc.)
- and about 100 more things that are getting more and more specific
All of this, of course needs to conform to brand guidelines and to platform guidelines and behaviours (the look and feel on smart TVs with navigation via remote will be different from mobile devices).
All of this, of course, needs to be retrieved from or sent to any number of internal Spotify services (even if some of them are not documented you can at least look at the source code) and partner-provided APIs (good luck finding docs for some of the devices).
There’s also A/B testing, metrics, experiments, white/balck-listing of clients models and platforms across various services for certain things etc.
And that’s just for skip button.
— end long story—
The absolute primitive stuff like “give me a media control strip using CSS” will definitely work, and will even give you working code (tried it with ChatGPT after I’ve already implemented it for an internal project). That you still need to spend several hours on because it conflicts with the rest of the UI and still needs to be hooked up to everything. Edit: Copilot was no better, on this project. It worked maybe twice, and only once it actually produced a valid result (which did save a few seconds of tedious work).
There will be no massacre.
i’m pretty sure this will be a major helper in the future, but right now it’s more trouble than it’s worth.
