Post

(codecompanion is definitely trying to be a lot, but i mostly just send current line or a selection to its chat buffer, neovim as input box.)

Surprised not to see anyone else mention Google's Antigravity in here. It's a vscode fork very similar to Cursor. I already paid for a bunch of Google storage so I get some AI for "free". I've been pretty happy with it though and have used Cursor, CC, Amp, and Zed in the past.

Getting good results with claude code in terminal with worktrees and for work that I'm directly driving zed with copilot is the only ide experience with valuable suggestions.

VS Code, Github Copilot with Claude Sonnet 4.6 and Claude Opus 4.6.

Claude Code is almost all via the terminal. OpenCode I use TUI mostly, and the web interface when I'm using worktrees

VS Studio Github Copilot and Github Copilot CLI, both with Claude Opus 4.6.

claude code. we run it on bedrock at work, and i have a personal subscription for myself. i literally learned neovim just because of claude code, but honestly these days i'm usually just viewing diffs in it (nvim is nice because it's fast to poke around various git worktrees with it).

Google Gemini "Thinking" (mainly in a "help me extend this function I wrote" way). [Doing little coding at the moment, but still more than I otherwise would be without some AI help]

Claude Code all the way, often through Conductor, which I've been falling a little out of love with, so I've made my own orchestrator UI to replace it.

You can replicate the basic idea by asking Claude to manually create worktrees and seed them with TASK.md.tmp files. Then manually spawning gemini's (with 3 pro, 2 is... bad) in those worktrees and tell them to read their task file, complete the task, and file a PR

And I usually have to do the inverse because Gemini is great at planning and reasoning, but Clause just works better as a collaborator in Copilot.

GitHub Copilot in CLI, VSCode and VS 2026

I also have it set to use Copilot as the model source, it turns out 3-5x cheaper for my usage pattern, even if I do miss the gigantic context windows when paying Anthropic directly.

I was mostly in Cursor and occasionally in VS Code’s terminal, but these days I keep it almost entirely in VS Code.

Claude Code.

Opus in Claude Code, with a dash of Pi on the side so I can tell CC to request external code reviews from Codex/Gemini/whatever via a Bash call. I currently work exclusively w/ Xcode, so I use a custom setup of docker container, host/guest MCP bridging, and worktree shenanigans.

For compliance reasons all of my professional work uses on-prem open-weights models. For coding I use cursor, other agent tasks use a custom framework.

Cursor for code (most recently had it make a wrapper for a free tool I use but doesn’t do exactly what I want. Gemini for non-code stuff (usually errors or how do I do X if a web search fails first, mostly ops stuff)

Now i obviously use 5.3-codex :)

The desktop app is less front and center, but it's there.

Claude Code primarily, but with Batty as a supervisor layer on top. It dispatches kanban tasks to Claude Code/Codex/Aider in parallel tmux panes, with test gates before merge. Going from one agent to supervised parallel agents was a game changer. github.com/battysh/batty

I use a mix of: Claude Code (Max 100 plan), OpenCode (with gpt-5.3-codex from OpenAI team plan), GitHub Copilot. Since the reliability and performance of Claude is a continuous up and down, it’s hard to use just a single product.

Codex.

Claude code plugin inside VSCode

Primarily Claude Opus 4.x via Copilot to assist with documentation and understanding of pre-existing codebases.

Cursor for now - mostly bc I signed up for a year, paid in full, last May. Will need to reevaluate if renewal vs Claude Code is the right call next.

Claude Code in VSCode and the CLI, depending

Zed with Anthropic models (still maining Claude Opus 4.5).

I thought gas town was kind of insane when I first read it but the podcast made the concept click for me. Now I think mass coding agent orchestration is the Next Big Thing ™️

At work a mix of the company's own and Claude Code. At home I daily drive Codex.

Claude Code

Claude inside VS Code

VS Code + GitHub Copilot + usually Claude Opus 4.5, sometimes Gemini 3 Pro.

Claude Code!

Claude Code usually with the VS Code extension, but sometimes just the CLI. Also often use Gemini CLI for reviews of complex plans.

I assume so.

It’s from these folks www.humanlayer.dev

Droid. Similar to Claude Code. Just appears more stable and reliable to me.

augment/intent, claude code, cursor of late, usually using opus 4.6 in all of them

Copilot CLI in Ghostty and VS Code for direct editing

any skills, prompts, plugins, etc?

Zed + zed agent, Claud agent, coded agent

@seldo.com I’ve been using Claude Code to great effect. Also if you like that command line app interface check out OpenCode which supports basically all the models and even gives you access to models via an existing Copilot subscription. Claude Code has a lot of whimsy in its interface too […]

I'm using Zed for the GUI. Planning on getting Claude Code soon. I might switch back to Cursor though.

Claude with a decision graphing and memory tool I made to give it nice guardrails and ability to recover state and do deeper analysis q&a super effectively to plan (the tool is agnostic tho it works with the major editors/opencode/claude) notactuallytreyanastasio.github.io/deciduous/

IntelliJ with the copilot extension running Claude sonnet 4.5 or 4.6

depends on what i’m doing, most of the time just directly in the CC terminal, otherwise zed+cc

cursor! been using it for two years

I am but I have yet to activate them.

Codex Mac app, Zed for manual editing on top (agents can’t write enterprise CSS)

This is how it looks on my XFCE Chicago95 themed session.

I should mention I've not yet evaluated codex or tidalwave. I also forgot to mention I used Aider for months. Its great, but opencode or claude code in the CLI is better.