I use lots of coding agents. I geek out on this stuff. Even run some LLMs on a server at home.
I'm not a developer, but I was a computer science minor. I'm pretty good with Python and C#. I use AI coding agents to build quick PoC (proof of concept) apps for my work. It's really insane what some of these things can do. They do make mistakes, and it's important to review the code (or even have AI review it, lint it, test it, etc.)
My preferred coding agent is Claude Code. I have a Pro subscription through my employer (it's $20/mo). I also use Gemini's free CLI coder from time to time. I hear good things about ChatGPTs recent codex agent, but I don't want to pay for another sub to try it out. In general coding agents are really well versed in the most common languages like Python, Javascript, C#, etc. However, I've found they're not so good on lesser used languages like C++. They're still great, but I suspect they get much less usage so there's less to train them on. My company built an internal coding agent that's trained on millions of lines of code in our codebases. It uses Claude as its foundational model.
I use CC in VS Code, but there are tons of other coding agents. Cline and Roo Code are really great for trying out different models and creating an agentic workflow. If you want to test some out, I'd recommend getting an OpenRouter account, put $5-10 in there, connect it to Cline/Roo and test some models (many are free). If you have a Gemini, Claude Pro, or ChatGPT paid subscription, you can get a lot of "free" use out of their CLI tools. They plugin really well to IDEs like VS Code, Cursor, etc.
It's fascinating (to me) to watch how fast things are evolving. The big three (OpenAI, Google, Anthropic) are constantly trying to one up each other.
Originally Posted by
KRSW
I don't use that, but we're in the process of automating Ms. KRSW's job via Python, using Claude AI to write the code. Their current systems were processing about 3-5 projects per hour. The little Python script Claude did for us in about 90 minutes of messing around with it can do 1 project every 10-30 seconds, and so far does a better job.
Neither one of us knows Python. My last programming gig was 30 years ago and that was in C, and I've not touched it since. We're learning Python so we can fix this thing if it breaks.
This is what's so cool about coding agents. You really don't have to know the language syntax, you just need to know how to manage the agents. The future of coding will be architecting, prompting and context engineering. Having said that, if you want to make production grade code, it is important to put the code in a proper review/test cycle (with a dev who can read/interpret it).