I’m curious to hear what kinds of AI coding tools people use and what you think about them.
So far I’ve used:
Copilot in VS Code, and on GitHub for code reviews.
Windsurf (referral link that gives some extra credits) — I think Windsurf is especially good for reading code and learning about it. It can write sidebar tutorials about small sections of code when you hover over names and click on “read more”.
Antigravity — I use this a lot, but it’s very aggressive about writing code, so it’s a little more difficult to ask it questions.
Codex
Gemini CLI
Claude Code — I’ve only used Claude Code for two days, and I’ve hit the usage limit twice in one day which has never happened with any of the other tools. I’m going to experiment with it more today.
Cursor (so far, just a 2-week trial a long time ago)
Ollama (not for coding, but for implementing LLM functionality in my own code)
I usually find myself using one of the Gemini, GPT, or Sonnet models. I’ve tried Opus a little but it burns through credits much faster.
I used the cheap plan for Claude Code and kept running out, so upgraded to Pro and still sometimes ran out, so upgraded to Max (months ago) and have not yet run out, mostly using Opus.
I also use opencode with the much, much cheaper Z.ai coding plan (GLM-4.7) which works well. I also use opencode with the Deepseek “pay as you go” API and only spend pennies per day–it’s pretty fast, too.
I also use Ollama and llama.cpp to run smaller and larger LLMs on local GPUs, but these run slower, and if I didn’t have grid-tied solar would be too expensive, given PG&E rates.
It seems to be the most expensive one. I haven’t run out on the other plans, but I regularly run out with Claude Code. I’m about to get locked out for 90 minutes more on the $20/month plan in the middle of something.
I still take the approach of spreading AI tasks over different ones. I sometimes use ChatGPT web chat to start a project that I want Claude Code to finish (downloading a zipped up project with build scripts, code, and documentation).
I sometimes use Gemini to research a topic and have a conversation about what to build before switching to Claude Clode.
I wrote a Rust CLI that uses local Ollama to scan through my github cloned repos locally, to analyze each one, summarize the purpose/state of each, suggest next steps, and recently I added the ability to suggest topics for creating YouTube videos.
Other ways to reduce Claude Code usage: gemini cli, codex, opencode with other cloud/local LLMs, and there is an interesting “code-diffusion” web chat that is free and very fast at generating code (to cut & paste).
Also, while waiting for the Claude Code lockout (before I upgraded), I just switched to other tools and other projects.
So it is an “LLM-agnostic” coding agent CLI that can be configured to use any cloud or local LLM. e.g., it supports both the metered Z.ai access to GLM-4.7 and the Z.ai coding plan (quota per day, subscription). It supports all the major cloud AIs and has a slightly different UI than the other CLI agents. It is open source (Go and Typescript IIRC), and very fast/capable. The advantage? It is not locked into one provider (though Claude Code, at least, can be configured to use other Cloud APIs)
