Comment by x1n13y84issmd42

I just don't see much use for AI coding agents beyond PoC and prototypes, that are to be ditched soon.

In my 15 years long developer career I've always found ways to automate all kinds of chores in big projects - simple shell scripts generate boilerplate code for services/modules/controllers/etc, modern infrastructure & tools automate or completely hide a lot of mundane work while giving back valuable data to act upon (think testing and deployment), the code itself is almost always highly modular and lets me to rearrange complex logic in no time with minimum risk. All that while having complete understanding of what's going on, and full control over every aspect of the software I'm responsible for.

And I have a feeling that writing proper prompts to achieve all that (even given the model has enough context) would simply be more work than just properly programming it the old way.

Besides that, the models feeding on OSS and Indian code from GitHub is not only injust, it's also very limited in how creative it can be - the next innovation would leave LLM-generated solutions behind if kept in vault, so LLMs cannot learn form it. And I have a feeling that's where we're heading - death of OSS (esp. in the wake of the recent supply-chain attacks) and a new round of closed-source technological competition.