I want to share my recent experience using advanced AI models for software development, and I would be very interested to hear from others in the Julia community to see whether I can improve my workflow. My experience so far has been poor, despite trying to follow the usual recommendations for maki…

Maybe JuliaHub can donate, or sell, high-quality datasets of numerical Julia programs to OpenAI and Anthropic to train GPT-6 and Claude-5. That would be the ultimate solution.

I find the cursor Debug mode with auto model selection already work well for me with data processing tasks without any skill configurations. Basic algorithms work. It’ s already great productivity boost for me. Haven’ t got a chance to try on advanced numerical methods like yours, though.

Yes, I feel that for data set exploration, even creating some GUI for visualization, these tools can help iterate a lot. But, after reading some of the “advances” people have been reporting over the last few months, I was expecting that, after spending some time refining the agentic workflow, it wo…

I find that agentic coding with Julia works well when you focus on boilerplate glue type tasks, but these systems struggle a decent amount reasoning about allocations, automatic differentiation, and using up to date best practices working with many packages. Using Augment Code w/ Sonnet 4.6 I put t…

I’ve only had success at generating API docs with AI. It’s also been useful to come up with partial solutions or to quickly discard avenues I was already suspicious of. It’s also competent at being a better autocomplete.

I’ve been using Claude CLI for about a month now and now am able to successfully implement large changes in codebases with it (e.g. adding database caching + logic for cache misses, AST evaluation, etc) by first drafting a PLAN.md (or similar document, the name doesn’t matter) with requirements, etc…

I’ve used codex (with GPT-5.4 high) quite successfully recently. But mostly as an interactive rubber duck to help with design ideas. I have used no AGENTS.md or anything like that. I’ve had most success by first “discuss” with it the design, once it has a narrow plan, it then is quite good at imple…

If you’re not using Kaimon.jl to work on Julia code with agents … well to be honest I think you’re crazy! :wink: I wouldn’t do it any other way! Kaimon.jl Opening the gate between AI and Julia

Thank you for pointing this out. I will definitely give it a try. Still, I am not sure if this will cover my desired use case, where the agent is smart enough to really recommend better numerical algorithms in different parts of my program, and code them in a way that do not bloat code. But the us…

Agent models often need to iterate on the code to get to good results. Providing the means to quickly and efficiently (with respect to time and tokens) do this iteration has provided huge benefits in my experiences. Also, use the best models (Opus 4.6 1M token context), use multiple models. Kaimon p…

Poor Impression on Agentic Development in Julia with Codex. Skill issue?

General Usage

mbauman Split this topic April 6, 2026, 5:59pm 15

A post was merged into an existing topic: [ANN] Kaimon.jl — opening the gate between AI agents and Julia

Topic		Replies	Views
Are there efforts to improve ChatGPT for Julia code? Tooling	26	4160	July 24, 2023
The use of Claude Code in SciML repos Tooling	26	6727	March 27, 2026
Vibe coding vs agentic coding (Ralph Wiggum loop) Offtopic	9	811	February 3, 2026
Looking for custom instructions / settings for coding agents General Usage agents , ai , llm	0	108	August 8, 2025
Reliability of AI coding tools Offtopic llm	31	721	November 19, 2025

Poor Impression on Agentic Development in Julia with Codex. Skill issue?

Related topics