“I've been using Claude Code a lot the last year, and I think it has been quite nice to use. I'm however looking into alternatives, as I feel like it can use unnecessary amounts of tokens for some tasks.”
“I keep seeing the same pattern in this sub: six months ago building any kind of AI agent felt wide open, now Claude Code and Codex are so capable that a lot of "build an agent that does X" ideas feel like they're one Anthropic/OpenAI release away from being absorbed into the base product.”
“You run Cursor you get one set of failures. You run Claude Code you get silence. Same repo, same standards, different outcomes depending on who opened the laptop. The inconsistency breaks CI.”
“Essentially, you open a repo in Claude Code - even with the sandbox enabled and the most strict permission mode - and using one short message, and a command from that repo ran on your actual Mac - completely outside the sandbox and without any permission prompt.”
“If you would use an IDE like vscode and the Claud plugin, wouldn’t you be able to use a git diff viewer (source control) and see the files touched, with the code lines highlighted? You can select those lines of code and send them directly to vscode Claude code plugin and ask questions or directly state file xyz line auto b?”
“I think Claude Code support images rendering, but I use Pi which I know also does it out of the box. With herdr you can even paste images directly over SSH”
“After some update around 2-3 weeks ago, this feature is gone, and I get only an asterisk in its place (which used to be what I get when the work is done, but now it's always there).”
“In my company they give each one claude code.The transition of winning an argument change to “claude said…” which are fucking annoying facing this shit in daily basis.”
“What worked for me running several scheduled Claude Code sessions a day on one repo: no long-lived conversations at all. Each session starts blank and reads a short entry checklist, then the state files: a ledger, a queue of open questions to the human with a deadline and a default answer, an append-only journal, a status file.”
“Switched most of my work over to Pi recently but the annoying part was all my chats lived in other agents. Months of Cursor and Claude Code sessions and I didnt just want to summarize them.”
“Then I use Claude Design for the UI stuff until I am happy, and then I use the "Send to Claude Code" function there to let Claude Code integrate it. It works pretty well. Same workflow as I mentioned earlier I open a WT and yada yada.”
“i remember being absolutely struck the first time i ever used claude code and it could just do things. i'm also a big fan of crush and pi.it's pretty early on.”
“Why run this way rather than entirely cloud-based sessions via Claude code’s interface on any device? I feel like I missing something about local machine sessions.”
“Lately starting to prefer Claude Code Desktop's Cloud environments over other options. Basically what's missing is the visual inter-agent coordination surface.”
“Fun fact: Claude Code can do this remote stuff out of the box. Just run `claude rc`, and it will start a lightweight service that connects to the cloud and let's you start sessions through [claude.ai](http://claude.ai) or the desktop app.”
“This was born from my long haul Claude Code development sessions and morphed into its current state from swapping around CLI services on the same projects.”
“Opus asking three times inside OpenCode but never inside Claude Code is the real headline here, and I wish the table was the whole post instead of the model rankings people are going to fixate on.”
“Claude Code complied every time but also said every time that the change contradicted the repo docs. Codex just said "done" with no further explanation.”
“What I meant was pane level status over real tmux CLI sessions, with Claude Code / Codex / etc still running as themselves. Same goal of seeing what is going on, different substrate.”
“I use claude code and cursor. Interesting that I use opus 5 in cirsor and claude code, but in cursor i wasted less tokens while i do some tasks. During auto-tests runs, Claude code re-watch tasks and context before start my auto-test.”
“I looked at a lot of other landing pages for inspiration, went back and forth with Claude Fable 5 in Claude Code to iterate on ideas, had it demo designs for individual sections using /design, then stitched everything together into a cohesive landing page.”
“The end result is amazing, workflows run much much faster and spend far fewer tokens than before. Claude pretty much wrote the whole thing itself, so I definitely recommend this as an experiment if you run dynamic workflows.”
“Working from one main session and distributing tasks through forks lets you work in parallel while keeping a centralized context across multiple sessions.”
“Problem i am facing is claude code interface is not too dopamine giving which my brain craves to consistently just using it for study. Are there any plugins/tools/skills anything else exists which can make the chat interface more visualizing and interesting?”
“I have 2 claude coded apps that serve me incredibly well. One for my job hunt and one for my online dnd sessions as a note taker. They alone are worth my subscription over the last 3 months.”
“I think I am officially impressed: 1. Vibe coding: May 2025 (Claude code) 2. Move 37: July 2025 (Kimi) 3. Real coding agent: December 6th, 2025 (Claude code) 4. Wow: September 5, 2026 (Astra)”
“Kind of bullshit, instead of cutting Claude Code token usage by 90% it is cutting Claude Code file read token usage by delegating reads to Gemini flash (with its own usage) and trusting the summary of files it provides is not crap.”
“it’s useful to separate Claude Code itself from the engineering practices around it. Anthropic has some great material for learning the tool, while the Udacity Claude Engineering Nanodegree puts more emphasis on the engineering side—loop control, guardrails, and building reliable workflows.”
“The workflow is to agree on an implementation plan that’s written to a file, let Claude code without oversight until it’s done, review the work against the plan file, use MCP to read Linear for the next ticket, and start the loop over. Watching Claude while it’s writing code is a waste of time.”
“If you are scared to run the shell command, just point claude code at the repo and ask it to make a code to do something similar. Star the repo if it was useful.”
“If you still fail to produce good, reliable, secure software with Claude code or codex after half a year I would recommend to strengthen your skills about how to specify and design software. You need to steer the AI.”
“Claude Code and Codex are extremely frustrating when you’ve never learned to exercise a bare minimum level of patience or self discipline. They’re truly terrible tools for tech-illiterate people who expect to push a button and have competence come out the other end.”
“Ive been delivering client projects, completely with claude code for the last several months. Ive built so many of my own projects from useful to absolute garbage.”