For me this is how all workflows with AI end up lookingif you really want robust products. Do whatever with AI, go fast (from the point of view of perception, initially) BUT it is not going to work without extra work.
and ho, the bloat, do not forget the bloat, which is technical debt towards the future.
I am not anti-ai per-se, but I consider the software I write as a prouct to add features to and maintain over time. So in this case I think it is not wise to say that bc you got something impressive fast you are done. Now you have bugs, architecture, bloat removal, and others...
Letting an AI manage all the workflow is a recipe for disaster in anything that is not strictly short-term. For this reason, I hardly code one-off scripts myself anymore and I hardly use AI for big things besides discussions with the prompt, reviews and snippets. For adding tests it can also be useful.
For full, long-term products, they try to sell agents, and tokens and the like. I think they do not work well enough what I tried. It always ends up as a bloated unmantainable mess.
Unless something that is totally autonomous (by this I mean 100% autonomous) and automated ever exists, I see writing software that can be maintained by humans still critical. As long as this exists, the productivity upper bound will be that of humans reviewing and driving the workflow, even if with AI support.
alt Hacker News
For me this is how all workflows with AI end up lookingif you really want robust products. Do whatever with AI, go fast (from the point of view of perception, initially) BUT it is not going to work without extra work.
and ho, the bloat, do not forget the bloat, which is technical debt towards the future.
I am not anti-ai per-se, but I consider the software I write as a prouct to add features to and maintain over time. So in this case I think it is not wise to say that bc you got something impressive fast you are done. Now you have bugs, architecture, bloat removal, and others...
Letting an AI manage all the workflow is a recipe for disaster in anything that is not strictly short-term. For this reason, I hardly code one-off scripts myself anymore and I hardly use AI for big things besides discussions with the prompt, reviews and snippets. For adding tests it can also be useful.
For full, long-term products, they try to sell agents, and tokens and the like. I think they do not work well enough what I tried. It always ends up as a bloated unmantainable mess.
Unless something that is totally autonomous (by this I mean 100% autonomous) and automated ever exists, I see writing software that can be maintained by humans still critical. As long as this exists, the productivity upper bound will be that of humans reviewing and driving the workflow, even if with AI support.