Neovim + Claude Code Outshines Agentic AI Coders
Ditched Conductor and Cursor for Neovim-based workflow with Claude Code: replicates parallel agents, handles 7-8/10 code review complexity natively via LSP, no app-switching needed.
Limitations of Agentic Coding Tools Exposed
The speaker, a former heavy Cursor user, shifted to agent-forward tools like Conductor for managing parallel AI agents on GitHub repos. Conductor creates isolated workspaces via git worktrees, runs custom setup scripts (e.g., conductor-setup.sh to migrate .env.local and seed data ignored by git), generates code with models like Claude Opus or GPT-4o, and provides a basic text diff viewer for reviews. PRs integrate GitHub status (CI/CD, checks), but the plain-text viewer lacks LSP features like jump-to-definition or find-references. For complex changes (>3-4/10 complexity), he'd switch to Cursor, breaking flow. This context-switching frustration—needing a full editor for deep dives—prompted reevaluation. Alternatives like Supermaven were tried but didn't resolve it. Decision: Rebuild in Neovim, leveraging its maturity for robust reviewing without new apps.
Quote: "the fact that I was always forced to open up another application when things got a little bit too complicated and it wasn't simple enough that I could just quickly review from the built-in text viewer here." (Highlights core pain point driving the switch.)
Replicating Workspaces and Agent Flows in Terminal
To match Conductor's workspace isolation, the speaker built 'yw' (Yorby Workspace) script—AI-generated via Claude Code. It fetches main, creates a git worktree/branch, runs conductor-setup.sh for env/seed migration, installs deps, then launches Claude Code. Usage: yw test-workspace spins up instantly. In terminals like Cmux (Ghostty-based, vertical tabs/workspaces with CLI alerts) or Warp, this enables parallel agents without browser dependency.
Neovim (Kickstart.nvim config by TJ DeVries) handles reviews: space gs for git status, space hd for horizontal diffs, full LSP (gd for definition, grr for references). Added diffview.nvim for VS Code-like file list + inline diffs (space do). Rarely writes code here—AI generates, human reviews deeply. Handles 7-8/10 complexity vs. Conductor's 3-4/10; Cursor only for 9-10/10 edge cases.
Tradeoffs: Vim learning curve (mitigated by Claude Code as "cheat sheet": "explain VS Code XYZ in Neovim"). No native multimodel UI, but CLI tools suffice.
Quote: "this is not that special. Like we can really replicate this using Neoim and T-Mox." (Underscores how agentic hype overlooks terminal power.)
Automating PR Monitoring and Merges
Conductor's PR panel (status polling, CI previews) replicated via Claude Code skills: 'babysit-pr' uses GitHub CLI to fetch PR, checks actions; auto-merges if green, fixes reds or pings human. Run via /loop cron (1-min intervals) for live updates. Matches Conductor exactly, but local/CLI.
Post-review: Commit, PR via CLI/AI, monitor in terminal. Full cycle: yw → Claude Code tasks → Neovim diffs → babysit-pr.
Quote: "if it's all green and ready to go, then merge it automatically. If anything becomes red... ping me." (Captures autonomous-yet-supervised agent handoff.)
Why Warp Terminal Edges Out Cmux and Others
Tested Cmux (Ghostty lib, vertical tabs, CLI alerts like "task done"), but lacked Conductor's inline commenting. Warp (sponsored, but earned switch) adds: Native Claude Code detection/alerts, built-in diff viewer with PR-style comments (send-to-agent button), tab configs for multi-pane setups (e.g., left: setup script, right: Claude Code—unblocks during installs).
Warp hybrids terminal + GUI polish: Lighter reviews without Neovim, comments pipe to agent seamlessly. Switched from Cmux as "pure terminal" couldn't match without heavy customization. Prior Warp churn (2021-2025) reversed for AI CLI first-class support.
Tradeoffs: Warp's AI features underused (prefers Claude/GPT CLIs); performance high via Rust.
Quote: "the one feature that really makes me like warp over CMUX right now is... built-in text viewer... you can also then leave comments... it automatically sends it directly into your cloud code agent." (Explains decisive UX win over terminals.)
Evolution from VS Code Era to Vim Revival
Background: 5-year Neovim hiatus for VS Code/Cursor (Vim bindings). AI boom circled back—old tech (Vim) + new (Claude Code) > flashy agents. Yorby AI monorepo (Next.js, Supabase) tests scalability. Humble pivot: Made fun of Vim users (4M-view short), now converts.
Results: Faster velocity, fewer context switches, scalable to complex codebases. Ramp-up tip: AI-query Neovim docs.
Quote: "I've been spending the past couple of weeks migrating my entire workflow to primarily using Neoim... having a neoimb based AI coding setup is kind of the best setup." (Core thesis after hands-on validation.)
Key Takeaways
- Build custom 'yw'-like scripts for git worktrees + setup to isolate AI workspaces, bypassing tool lock-in.
- Use Neovim (Kickstart + diffview.nvim) for LSP-powered reviews: gd/grr beats plain diffs for 7-8/10 complexity.
- Claude Code skills (/loop babysit-pr) automate PRs: Poll GitHub CLI, auto-fix/merge, alert on fails.
- Prefer Warp for hybrid terminal: Inline comments-to-agent, multi-pane tabs unblock workflows.
- AI as cheat sheet accelerates Vim ramp-up: Query "VS Code action in Neovim".
- Rank tools by review depth: Neovim > Warp viewer > Conductor; reserve Cursor for extremes.
- Replicate agentic flows in terminals—hype tools add friction for pros.