Which AI Coding agent is best at writing an automation tool?

[vlringmo]

Claude, Codex or Gemini or what else?
Which one is better at coding for the daily routine download links from websites and uploads then posts to somewhere automation tools?

 

[Casino321]

Claude is better it give best response and exact codes, but it need subscriptions which r costly

 

[Northern Lights]

the $100 plan of Codex is a great value for price. Gpt 5.6 on extra high is super capable and the limits are enough to build real stuff.

 

[magnus1]

Claude Code and OpenAI Codex

 

[jackdaniel]

claude + python is exactly what you need

 

[permendrain]

Try Neoncodex.io , they provide all the latest AI Models (Claude, Chatgpt, Deepseek, GLM, Kimi, Nimitron and many more) in $29.

 

[daxmercer]

I've been using Cursor, Codex CLI, and Codex Cloud. That combo has been working really well for me so far.

 

[arnieson]

I tried Claude Code, OpenAI Codex and Gemini. All with a couple of simple projects like spreadsheets and python coding, and Claude Code has consistently given me better results.

For automation I suggest playwright with Claude Code. If you install the playwright Chrome extension it can even use your open sessions.

 

[Casino321]

I suggest cloude AI but its a paid ones and not cheap
Second suggest Deep seek or chatgpt

 

[NiceNIC_COM]

Claude or Codex can both help, but the tool design matters more. These automations usually break when a login expires, a site changes layout, or one upload fails.

 

[fengfengfeng]

cursor

 

[hexmy]

Claude, Codex or Gemini or what else?
Which one is better at coding for the daily routine download links from websites and uploads then posts to somewhere automation tools?

I’d say Antigravity AI is worth trying for this.

It’s pretty good for building automation tools because you can describe what you want in a detailed prompt and let it generate the code, then iterate on the implementation with follow-up prompts.

If you’re in India, there’s also a nice bonus — Jio users can currently get an 18-month subscription offer, so it can be a good option to try without spending much upfront.

For automation projects, I’d definitely give it a shot before deciding which coding agent works best for you.

 

[vvxx]

Claude AI

 

[RoRoNo]

I’ve tried Claude, Codex, and GLM Flash.
All the models worked well.
But GLM Flash is cheap.

 

[1001SMS]

I’ve had the best results with Claude for planning/refactoring and Codex for implementation. For automation work, I’d test both on one real workflow and compare reliability, debugging, and how often they need manual correction.

 

[majorphones]

for a download/upload/post bot like that the model matters less than making it log every step and retry, it will break every time a host changes something. also have it keep the selectors and endpoints in one config file, then fixing a layout change takes 2 min instead of a rewrite

 

[Orenix]

I would add one specific test to the logging/retry advice: interrupt the workflow after the upload succeeds but before the forum post is recorded as complete. On restart, does it reuse the uploaded file and avoid publishing twice?

Give each coding agent the same small workflow and three failure cases: an expired session, a successful upload followed by a timeout, and a page where the expected button is missing. Ask for a job log containing the input ID, completed stage, uploaded-file ID and published-post ID. That gives you something concrete to compare.

For context, I work on Orenix, which turns crypto trading-signal text into structured setups. That makes reproducible inputs and explicit failure states especially relevant to me. For your automation, I would judge the generated tool by how safely it resumes a failed job, not just whether the first run works.

 

[ITJobsInPakistan]

I agree with the logging and retry point, but for a download-upload-post workflow I would split it into three separate stages with a checkpoint after each one. Download, upload, and post should not run as one long job, because if the post fails you should not have to download and upload again.

Before posting, the tool should check whether that uploaded file ID has already been posted. If it has, it should skip the post instead of publishing it twice. I would also keep a simple record of the original link, the downloaded file, the uploaded file ID, and the post ID, so a restart picks up from the last completed stage, not from the start. For a daily routine job like this, that separation matters more to me than which model wrote the first version of the code, because the failures will mostly come from expired sessions and changed pages, not from the code itself.

 

[voreu]

Hello.
Claude +1