Raw LLM Responses

Been coding since I was a kid and then 30 years of professional coding. I love doing it, too! Started playing around with Github Copilot in my VS Code way way back when GPT-4 first came out, heh. I think it was in 2022. Found it to be semi-tarded and uninstalled that extension. A year later, it improved and I learned to use it but with great care and many git commits and reverts. Between then and now, I've seen steady increases in quality (usually in the underlying models, yay Claude 3.5 and now 3.7!). Finally, now there are quite a few high quality competitors for Github Copilot that in my opinion, blow it out of the water. My current love affair is with Cline (free), one of the newer "agentic" code assist extensions. I love the variety of underlying models I can tell it to use. Just found if I point it at OpenRouter as API provider, I can access Deepseek's currently free v3, which is close to Claude 3.7 in its intelligence and yeah, FREE. Right now, I switch to Claude 3.7 and turn on "enable extended thinking" for the hard stuff. Also, yeah, you can toggle "auto approve" on/off and just watch it go with coming up with a plan of action. Once the plan is complete, you can tell it to modify the plan and/or just click "Act" and watch it go make changes across all the files in your code base it needs to change. With the right underlying model, it can do a lot! And sure, yes, I still have to babysit and correct things often. But it is a huge improvement on Github Copilot. It's basically a free version of Cursor. OK so here we are now in March of 2025 and these code assistants and LLM models are as dumb as they will ever be. You make a great argument for coding not going away and actually increasing. You seem to be assuming, though, that these assistants and LLMs (and what is coming post-LLM) will not improve. Smarter, larger context windows, etc. Right NOW they nail small applications. Have you noticed how rapidly new models are coming out that leapfrog the capabilities of old ones? Extrapolate. We are seeing significant jumps at least every month! I predict that within 2 months, the size of a robust app that a coding agent (notice I've changed from calling them assistants?) will have doubled. And I predict that size will double again and again every couple months, on average. The writing is on the wall. This is not like when people feared the dominance of CMSs would put software engineers out of jobs. Just around the corner is the arrival of a new kind of software engineer that is cheaper, more emotionally resilient, and more physically resilient than a human. And yes, very very soon, more skilled than any human. Advice: Learn to code for the self-discipline, mental stimulation, leveling up your problem-solving skills, and general growth, not for future money/success. Instead, learn how to use and communicate with LLMs. Sadly, it's a moving target. For example, prompting techniques right now will be obsolete in mere months as LLMs (and their successors) get so smart, it doesn't matter how you phrase your request. The future is scary and the future is so bright I gotta wear shades (~ Tom Petty).