alt Hacker News> The result is a lot of proof-of-concept projects that lack what's required to make them durable, trustworthy, and deployable at scale. Starting with business needs first is essential.
A bunch of frivolous projects that fail sounds to me like a pretty good way to learn how far a new technology can be trusted.
If you're considering putting AI into something load bearing you either need a engineer who has not been participating so they can say "no" or one who has made 15 failed AI projects so they can say "maybe". The very worst case is to pressure somebody who doesn't know the technology very well into saying "yes".
Replies
> ...learn how far a new technology can be trusted
> Starting with business needs first is essential
But the reality is that for many end consumers of software, it's not a good experience to use janky software that changes behaviors, flows, and screens on a whim now because product can.
I think AI has had a negative effect on product teams; I can see all pretense of thoughtful design and execution after understanding the customer being thrown out the window and leaving a much worse end-user experience as designs and capabilities shift around without foresight and product teams "feel" their way through.
[flagged]
The management is the team that wants it to go, so they do not enable anyone to say no.
I have been in meetings where the top item is ensuring security of XYZ llm component, and after we've shown its inherently not something secureable from what the product requirements are then those product requirements are discarded.
For many of these companies the entire thing is a smoke and mirrors game to just get more money, they have little to no commitment to ... anything really.