@iannuttall I like to try all of them and end up building nothing

@petergyang hazzah, a man of culture!

@iannuttall - @AmpCode - @vibekanban (naturally) - VS Code @AmpCode Oracle has gotten really good for planning. Pretty sure it’s GPT-5 by default now. Not quick or cheap, but find it spots and re-uses patterns more often.

@LouisKnightWebb @AmpCode @vibekanban yeah, saw they switched it the other day!

@iannuttall Is Codex legit? Are people really switched over from Claude Code??

@anthony_harley1 it's very barebones, the ui is minimal which i personally love, and it relies a lot on just the gpt-5 models being really good! but i'm not sure it's as good of an agent as cc or factory

@iannuttall i’m still a beginner but using cursor with claude 4.5 has been a game changer. it helps me debug faster, make quick front end tweaks, and experiment with new web features in a controlled environment that’s easy to move around in.

@iannuttall - pycharm/vs code + copilot - I used pycharm before and I got used to it. So I just added Copilot - I'm a dev but now I usually ask Copilot to do stuff and then just check and tweak the result. Copilot + gpt5 usually takes up to 2 minutes to do its job. During which I can

@iannuttall cursor w/ GPT 5 high to look at my code base and come up with implementation instructions, Gemini 2.5 pro (aistudio), holder of context. Opus 4.1 (claude code), executes instructions given by GPT5. For brainstorm, I often use Grok 4 expert.

@iannuttall Codex with spec-kit, it is 🧑‍🍳s💋 Guard rails that actually work with iterative dev process baked-in Vibing on the weekends, proffesh tech lead by day but plan to introduce it to my team (will be restricted to copilot CLI but honestly don't think the agentic tool-use loop matters

@iannuttall Codex CLI for now It’s minimal and precise and doesn’t go tryhard mode like Claude doing unnecessary changes. Code generation isnt my bottleneck, ideas and clearly defined problems are. I dont love hopping between coding tools. I like the dev experience with codex so im stickibg

@iannuttall It may sound weird but I am using a combination of lovable and cursor! Lovable was the starting point for us and cursor is enhancing our code base! And I am a vibe coder who is learning professional dev through vibe coding!

@iannuttall Been slaying 10k clips of code with OpenCode using a combination of Kimi K2 and Grok Code. The agentic framework I cobbled together seems to do the best with Kimi in charge and Grok doing the lifting in the form of spawn agents. Not a vibe coder, not a professional dev. I guess

@iannuttall Droid with GLM 4.6 and special automation rules for initial codebase setup with minimal intervention by me. Runs for hours. Setup based on very detailed PBIs that droid generates according to a PRD that I provide. The agentic system with many sub-agents also tests everything

@iannuttall pro dev using cursor. I've tried other stuff like Gemini CLI and Clide inside vscode but nothing seems to have anything like cursor's tab autocomplete.

@iannuttall Tried Cursor for a year, back to VS Code now. Current stack – 1. Copilot for autocompletion 2. Codex inside VS Code for most tasks 3. Claude desktop for planning/ideation/base code. also a custom script that generates context from selected files, paste into Claude code...

@iannuttall Cant get away from claude code even though codex and gpt-5 seem to be more accurate. I like sonnet 4.5 for speed/performance and clarity in what it does. With good prompting and context it can go a long way. Vibe coder and not professional dev

@iannuttall Claude code for dev work Cursor for QA Rotating between multiple for Tech Lead / Architect (most opus and Gemini these days) All of them converse with each other and use some central process files for planning awareness core docs.

@iannuttall I’m using Claude code mostly but I have sub agents that collaborate with Codex. Also using Conductor, but for long running autonomous coding, I need verification using iOS simulator using MCP which is painful with multiple instances in work trees.

@iannuttall what: headless claude code via repoprompt, sometimes paired with cursor, codex, amp or factory the paired agent coordinating and the headless claude code executing via repoprompt mcp why: repoprompt allows for optimal context engineering and agent flexibility i'm a pro dev

@iannuttall I've been testing @FactoryAI recently, and I'm generally impressed with the results. However, it still depends on the prompt and the PRD quality. Claude Code Sonnet 4.5 with @RepoPrompt uses GPT-5 Codex and does a great job. I'm still experimenting. Doing Pro dev on a large

@iannuttall Claude Code: SOTA intelligence + most easy to follow what is being written due to well designed UX

@iannuttall Claude Code INSIDE @boltdotnew Why: I like having the infra integrated for backend & deploy (I also work there) I’m definitely a vibe coder, but can review simple code

@iannuttall Happy Coder with Codex for planning out changes on my morning walks (using the Happy voice agent). Then I continue the same session at my laptop, and review diffs and edit with Cursor.

@iannuttall Claude code just feels better than cursor with same model imo. I trust Anthropic’s stack and context handling more overall. Codex rn feels too slow and not that much better although if Claude struggles w something I’ll test codex.

@iannuttall Claude Code - The SDK is unmatched for building custom agents - The subscription offers exponentially more value in tokens than you pay for - Works on CI - Override the system prompt w/ outputStyle - Most robust hooks system

@iannuttall Cursor or VS Studio as IDE then mainly Claude Code and switch to Codex if CC gets stuck. Also Factory is solid. I need the IDE for docs (context management), manual code edits (normally copy edits!) and just like to have visibility on the directory. Used to be FED 20 yrs ago.

@iannuttall I tried all of them and ended up with Replit. It's best for casual dev who wants to put something out in prod without dealing with devops and setting up local env.

@iannuttall Lovable Simply, Lovable can dev frontend, backend n design consistently n shortly, I’ve been using codex, cc, cursor though. I’m vibe coder. Strongly recommend: - Lovable if you wanna ship shortly in a few days - codex or whatever if you wanna dev like professional

@iannuttall -Just swapped back to Copilot (the $39 plan, from CC Max plan) . -New models, agents are pretty good, I missed my VS code extensions, I prefer to use AI for individual programming tasks, not complete major features at it. Cost was a minor factor, more value for $. -Pro

@iannuttall Okay, this is tough because the AI tools are! But honestly, my daily driver is still VS Code + Claude code. I'm a professional dev who tries to maintain a vibe coder spirit, gotta have fun while shipping! Don't chase every shiny tool, master one.

@iannuttall - paste diffs between claude code and codex cli - more than hallucinations, reward hacking and half-way implementations / magic numbers / hard-coding things is very tough to circumvent - especially as a vibecoder - so i paste back and fourth until there's a SSOT that's

@iannuttall I’m using only Cursor lately—flexible, lightning fast, and lets me create from idea to implementation, real quick. Definitely a vibe coder!

@iannuttall Augment Code. They’ve got very good plugins for the top IDEs I use ( especially IntelliJ, Android Studio) for a Java Dev like me. But, I have recently discovered Droid (by Factory) from one of your posts I believe, and honestly I’m deeply impressed by its depth of code

@iannuttall Used the max plan of CC for nearly three months, after which I switched to Codex for a brief period. Now using droid (saw it first on your account) and switching between sonnet 4.5 and gpt-5-codex. Had a lot of fun with it today, it's a really solid cli agent

@iannuttall codex cli lately because i like controlling the environment completely and the gpt-5-high reasoning is noticeably better for architecture, professional dev who codes manually sometimes but leans heavily on ai now