My AI Dev Workflow in 2026

Every developer's setup is different, but in 2026, the ones shipping the fastest share a common pattern: terminal-first, AI-centered, and browser-optional. This is not a theoretical guide. This is the exact setup I use every day, the tools I have tried and abandoned, and the configuration files that make it all work.

If you are still switching between your editor, a browser with twelve documentation tabs, and a separate chat window for AI assistance, you are leaving speed on the table. The goal is to keep your hands on the keyboard and your context in one place.

Key Takeaways

• A three-pane terminal layout with Claude Code in the center pane eliminates context switching between AI, code, and output
• Git worktrees replace git branches for parallel feature work without stashing or switching
• Claude Code's /add command replaces browser-based documentation lookups by pulling relevant files directly into context
• Shell aliases and custom scripts reduce multi-step operations to single commands, compounding time savings across hundreds of daily operations
• The editor becomes a viewer, not a writer -- you review and approve changes rather than typing them character by character

The Terminal Layout

I use a tiling terminal (iTerm2 on macOS, but tmux or Wezterm work identically). The layout is three vertical panes.

Left pane: Project context. This runs git status, git log, file watchers, and test runners. It is my dashboard for understanding the current state of the project.

Center pane: Claude Code. This is where the work happens. Claude Code runs here in interactive mode, and I spend 80% of my time in this pane.

Right pane: Server and output. Dev server logs, build output, and occasionally a second Claude Code session for a different task.

# iTerm2 profile setup (saved as "AI Dev" profile)
# Three vertical panes, center pane is 50% width
# Left: 25% | Center: 50% | Right: 25%

The key insight is that the center pane is intentionally the widest. Claude Code output benefits from horizontal space because code snippets and diffs render better with wide lines. If you want to go deeper on multi-session management, see our guide on managing multiple Claude Code sessions.

The Editor Question

This is controversial: I barely use my editor for writing code anymore. I use VS Code as a file viewer. I open files to review changes Claude Code made, occasionally to make a quick manual edit, but the days of sitting in an editor typing line by line are over for most tasks.

That said, VS Code still earns its place in the workflow for three reasons.

Diff review. The built-in git diff viewer is better than terminal-based diff tools for understanding changes across multiple files.

Search. VS Code's fuzzy file finder and global search are faster than grep for navigating unfamiliar codebases.

Extensions. Error Lens, GitLens, and the Tailwind CSS IntelliSense extension provide visual feedback that complements Claude Code's text-based output.

Editor Configuration

{
  "editor.fontSize": 14,
  "editor.minimap.enabled": false,
  "editor.wordWrap": "on",
  "workbench.activityBar.location": "hidden",
  "workbench.statusBar.visible": true,
  "editor.renderWhitespace": "none",
  "breadcrumbs.enabled": false,
  "editor.glyphMargin": false,
  "editor.folding": false
}

Minimal UI. Maximum code visibility. Every pixel of chrome you remove is a pixel of code you can see.

Git Worktrees Over Branches

I stopped using git checkout to switch branches a year ago. Git worktrees let you have multiple branches checked out simultaneously in separate directories. Each worktree is a full working copy with its own working tree and index.

# Create a worktree for a new feature
git worktree add ../worktrees/feature-search -b feature/search

# Each worktree gets its own Claude Code session
cd ../worktrees/feature-search && claude

The advantage is zero context switching. Your main branch stays clean in your primary directory. Your feature work lives in a separate directory. You can have Claude Code running in both simultaneously, working on different features in parallel.

When the feature is done, merge and clean up:

git worktree remove ../worktrees/feature-search

Shell Aliases That Save Hours

Small automations compound. Here are the aliases that save me the most time.

# Quick Claude Code with project context
alias cc="claude"
alias ccr="claude --resume"
alias ccp="claude --print"

# Git shortcuts
alias gs="git status"
alias gd="git diff"
alias gl="git log --oneline -20"
alias gw="git worktree list"

# Project navigation
alias proj="cd ~/AI/AI_Coding/Repositories"

# Quick file creation with boilerplate
alias newcomp="claude 'Create a new React component in components/ with proper TypeScript types and Shadcn UI imports'"

The ccr alias is particularly valuable. Claude Code's --resume flag picks up the last conversation, which means you can close your terminal, reopen it, and continue exactly where you left off with full context.

The CLAUDE.md File

Every project I work on has a CLAUDE.md file in the root. This file tells Claude Code about the project's conventions, tech stack, and patterns. It is the single most impactful thing you can do to improve Claude Code's output quality.

# Project Instructions

## Tech Stack
- Next.js 15 with App Router
- Tailwind CSS + Shadcn UI
- Supabase for database
- TypeScript strict mode

## Conventions
- Server Components by default
- Server Actions for mutations
- Use cn() for conditional classes
- All components in /components with PascalCase names

## Do Not
- Use Pages Router
- Use Material UI or other component libraries
- Use useEffect for data fetching

Claude reads this file automatically and follows the instructions. No more repeating yourself across prompts. No more Claude suggesting Express when you are building a Next.js app. For more on configuring Claude Code, see the CLI commands reference.

Daily Workflow in Practice

Here is what a typical development day looks like.

8:00 AM -- Morning standup with myself. Review yesterday's commits, check open PRs, identify today's priorities. This takes 5 minutes in the left terminal pane.

gl  # git log, see yesterday's work
gs  # git status, see any uncommitted changes
gh pr list  # check open PRs

8:10 AM -- Feature work begins. Create a worktree if needed, start Claude Code, and begin prompting.

# If starting a new feature
git worktree add ../worktrees/feature-filters -b feature/filters
cd ../worktrees/feature-filters
claude

8:15 AM to 12:00 PM -- Build, review, iterate. The core loop: describe what I want, watch Claude build it, review the diff, run tests, commit, move to the next thing.

12:00 PM -- Commit and push. Everything built this morning gets committed with descriptive messages and pushed up for review.

1:00 PM to 5:00 PM -- Continue feature work or tackle technical debt. Afternoons are often for refactoring, performance optimization, or documentation -- tasks where Claude Code shines because they are tedious for humans but straightforward for AI.

5:00 PM -- End of day review. Run the full test suite, review all diffs from the day, make sure nothing is broken.

Tools I Tried and Dropped

GitHub Copilot. I used Copilot for two years. It is good at line-level completion but cannot handle multi-file changes or understand project-wide context. Claude Code replaced it entirely.

ChatGPT for coding questions. Switching to a browser, pasting code, getting an answer, switching back, pasting the answer -- this workflow has too many context switches. Claude Code answers the same questions without leaving the terminal.

Notion for technical documentation. Switched to markdown files in the repo. Claude Code can read and update them. Notion cannot be part of the development loop.

Docker for local development. For most web projects, running the dev server directly is faster and simpler. I still use Docker for databases and services that need specific versions, but not for the application itself.

The Setup Checklist

If you want to replicate this workflow, here is the minimum viable setup.

1. Install Claude Code: npm install -g @anthropic-ai/claude-code
2. Set up a tiling terminal (iTerm2, tmux, or Wezterm)
3. Create CLAUDE.md in your project root
4. Add the shell aliases from this article to your .zshrc
5. Start using git worktrees instead of branch switching
6. Install the first skill to extend Claude Code's capabilities

The entire setup takes about 30 minutes. The productivity gains start on day one.

FAQ

Do I need a powerful machine to run Claude Code?

No. Claude Code sends your prompts to Anthropic's API and streams responses back. The compute happens on Anthropic's servers. Any machine that can run a terminal and a dev server is sufficient.

How much does this workflow cost per month?

Claude Pro at $20/month covers most individual developer usage. Heavy users may hit rate limits and need Claude Max at $100/month or $200/month for higher limits. The ROI is positive for anyone shipping code professionally.

Can I use this workflow with languages other than TypeScript?

Absolutely. Claude Code works with any language. The terminal-centric workflow is language-agnostic. I have used the same setup for Python, Go, and Rust projects with minor adjustments to the CLAUDE.md file.

What about pair programming with another human?

This workflow is designed for solo development. For pair programming, I share my screen and we collaborate through the Claude Code terminal. One person prompts, the other reviews. It works surprisingly well.

How do I handle secrets and environment variables?

Keep .env files out of git (add them to .gitignore). When using worktrees, copy .env files manually or use a script. Never paste secrets into Claude Code prompts -- they go through the API.

The Compound Effect

This workflow is not about any single tool or configuration. It is about eliminating friction at every step so that the compound effect of hundreds of small optimizations adds up to a fundamentally different development experience. Each alias saves 3 seconds. Each worktree saves 5 minutes of context switching. Each CLAUDE.md instruction prevents 10 minutes of re-explaining.

Over a week, these savings add up to hours. Over a month, they add up to days. Over a year, they are the difference between shipping 12 features and shipping 50.

Explore production-ready AI skills at aiskill.market/browse or submit your own skill to the marketplace.

Sources

• Claude Code Official Documentation - Setup and configuration reference
• Git Worktrees Documentation - Official git worktree reference
• iTerm2 Documentation - Terminal configuration guide