The best engineers don’t open ChatGPT in a tab anymore. Here’s what they do instead. Image generated by Author using Leonardo AI There is a quiet shift happening inside the best engineering teams right now, and it has nothing to do with switching languages or frameworks. It is about where AI lives in a developer’s day. The developers who are truly pulling ahead in 2026 are not the ones with a Claude tab pinned next to their code editor. They are the ones who have pushed AI directly into the terminal, into the git workflow, into the CI pipeline. AI is no longer a tool they switch to. It is a tool that is already there, waiting inside the environment where the actual work happens. As of early 2026, roughly 4% of all public GitHub commits are authored with AI assistance through terminal-based tools. Claude Code alone went from 17.7 million daily installs to 29 million in just a few months. These are not hobbyists. These are production engineers at serious companies, shipping real code with AI running alongside them in the terminal. This article is not about AI theory. It is a practical breakdown of five terminal workflows that actually change how fast you can build, debug, and ship. Each one is something you can set up today. Why the Terminal? Why Not Just Use a Chat Window? The honest answer is context switching cost. Every time you copy code into claude.ai, paste in an error message, read a response, copy it back, and then figure out where to put it, you lose a little bit of flow. It is a small tax per interaction, but across a full day of development it adds up to something real. The terminal is where your code actually runs. It is where your git history lives, where your tests execute, where your build fails. When AI lives there too, the feedback loop gets much tighter. You describe a problem, the AI reads your actual files, makes the change, runs your tests, and tells you what happened. You never leave the environment. That is the core insight behind every workflow in this article. Workflow 1: The CLAUDE.md Memory System (Stop Repeating Context Every Session) One of the most underused features in terminal-based AI development is persistent context. Most developers start a new session and spend the first few minutes re-explaining their stack, their conventions, their folder structure. That is wasted time. Claude Code solves this with a file called CLAUDE.md, a markdown file you add to your project root. Claude reads it at the start of every session. Automatically. No prompting required. Here is a practical example of what to put in it: # Project: PaymentsAPI ## Stack Node.js 20, Express, PostgreSQL, Prisma ORM, Jest for testing ## Conventions - All API routes go in /src/routes - Business logic goes in /src/services, never in the route handlers - Use async/await, never callbacks - Write tests before writing implementation - Commit messages follow: type(scope): description ## Things to avoid - Never use any as a TypeScript type - Never write raw SQL, use Prisma queries - Do not add console.log statements to production files Now every session starts with Claude already knowing your project. It will not suggest raw SQL when you use Prisma. It will not put logic in your route handlers. It will not generate JavaScript when you are working in TypeScript. The real power comes when you build CLAUDE.md files at multiple levels. Put a global one in ~/.claude/ for your personal preferences across all projects (preferred testing style, code style, languages you work in). Put a project-specific one in each repo for that project's conventions. Claude reads both. Teams shipping the most consistently in 2026 treat CLAUDE.md the way they treat a good README: it is living documentation, updated whenever a new convention is agreed on. Time to set up: About 15 minutes. Time saved: Significant, especially across a multi-person team where everyone’s Claude sessions behave consistently. Workflow 2: AI-Driven Git (Let the Machine Write Commit Messages and PRs) This one sounds small until you realize how much time and mental energy goes into writing commit messages, PR descriptions, and changelogs across a week. Claude Code integrates directly with git. From your terminal, you can do this: claude "commit my changes with a descriptive message" Claude reads the staged diff, understands what changed and why based on the code, and writes a commit message that actually describes the work. Not fix stuff or update component. An accurate, conventional commit message. But the real workflow upgrade is pull request generation. After you finish a feature branch, instead of manually writing a PR description: claude "create a pull request description for these changes, including what changed, why, and how to test it" Claude reads your entire branch diff and produces a PR description that covers the technical changes, the reasoning, and the test steps. The kind of PR description that makes reviewers happy and gets merged faster. You can extend this further into your CI pipeline. Teams using Claude Code in GitHub Actions are running automated code review on every PR before a human reviewer ever looks at it. Claude flags anti-patterns, checks for security issues, and generates inline documentation as part of the automated build process. This is not a developer using a tool. This is AI running as a component inside the infrastructure. Practical setup: Install Claude Code, then add a GitHub Actions step that calls claude with a review prompt on each PR open event. Anthropic's documentation has a working example for this. Workflow 3: The Split-and-Merge Agent Pattern (Parallelize Work You Used to Do Sequentially) This is the workflow that genuinely changes the productivity ceiling, and most developers do not know it exists. Claude Code supports spawning multiple agents that work on different parts of a task simultaneously. One lead agent coordinates the work, assigns subtasks, and merges results. This is called the split-and-merge pattern. A practical example: you need to add authentication to a web app. The work involves backend route changes, middleware, database schema updates, frontend form changes, and tests for all of the above. Traditionally you do these sequentially, or split them across your team. With multi-agent Claude Code: claude "implement JWT authentication. Split this into parallel tasks: (1) backend routes and middleware, (2) database schema and Prisma migration, (3) frontend login and registration forms, (4) test coverage for all of the above. Coordinate and merge." Claude fans out up to 10 subagents working simultaneously, each handling a scoped piece of the task, with the lead agent checking for conflicts and merging the results into coherent, working code. The key to making this work well is task isolation. Define the boundaries clearly before you let the agents loose. When tasks overlap too much, the merge step gets messy. When they are cleanly scoped, this pattern can do in 20 minutes what would normally take a full day of sequential development. This is the workflow that is making senior engineers look like entire teams. It is not magic. It is just smart task decomposition paired with AI that can actually execute in parallel. Workflow 4: Custom Commands (Automate the 15 Things You Do Every Day) Every developer has a set of tasks they repeat constantly. Generating a new component with boilerplate, setting up a test file, writing a feature spec, reviewing a specific piece of code for security. These are the tasks that individually take five minutes but collectively eat your day. Claude Code lets you build custom slash commands for these. Create a markdown file inside .claude/commands/ in your project (or ~/.claude/commands/ for global commands) and you have a reusable prompt template that runs with a single command. Here are three worth building immediately: New component command saved at .claude/commands/new-component.md: Create a new React component called $ARGUMENTS. Include: - TypeScript interface for props - Styled-components setup matching our existing theme - A basic unit test file - Export it from the components index Follow the conventions in CLAUDE.md. Now you run /new-component UserProfileCard and get everything scaffolded correctly, every time. Security review command saved at .claude/commands/security-review.md: Review the following code for security issues only. Focus on: input validation, SQL injection risk, authentication bypasses, exposed secrets, unsafe dependencies. Rate each issue by severity: Critical, High, Medium, Low. Do not comment on style or performance. Refactor plan command saved at .claude/commands/refactor-plan.md: Analyze the current file for code smells and refactoring opportunities. Identify: duplication, overly complex functions, poor naming, missing abstractions. Propose specific changes with risk assessment for each. Do not make any changes yet. Just produce the plan. That last point matters. Building commands that plan before executing is one of the most underrated practices in AI-assisted development. Getting Claude to show you the plan and confirm before touching files means you catch bad ideas before they touch your codebase. Workflow 5: The Warp AI Shell Workflow (AI That Understands Your Entire Terminal History) Warp is a terminal that has built AI natively into the shell experience in a way that other tools have not caught up to yet. If you are on macOS and you have not switched, this one is worth knowing about. What makes Warp different from just running Claude Code in a regular terminal is that Warp’s AI understands your entire session history as context. It sees every command you ran, every output you got, every error message. When you ask it a question, it is not answering in isolation. It is answering based on what you have actually been doing in this terminal session. Practical workflow: You are running a deployment and something fails with a cryptic AWS error. Instead of copying the error, opening a browser, and searching, you just press Ctrl+~ inside Warp to open the AI input. Type “what went wrong and how do I fix it.” Warp reads the last N commands and their outputs, understands the deployment context, and gives you a targeted fix. This is a fundamentally different interaction than pasting an error into claude.ai, because claude.ai does not know what command you ran before that error, what environment variables are set, or what the successful output looked like two steps earlier. Warp does. Warp also lets you run natural language commands directly in the shell. If you cannot remember the exact ffmpeg command to convert a video to a specific format, you just describe what you want and Warp translates it into the correct shell command, shows it to you for confirmation, then runs it. The confirmation step is important. Never let AI run shell commands without showing you what it is about to execute. Good terminal AI workflows are always one confirmation step away from the actual execution for anything that writes, deletes, or sends. The Mindset Shift Behind All of These There is a pattern running through all five of these workflows that is worth naming explicitly. The developers getting the most out of AI in the terminal are not using it as an answer machine. They are using it as an execution environment for well-defined tasks. The difference is significant. An answer machine gives you text you have to go implement. An execution environment takes a scoped task, reads your actual codebase, makes the changes, runs the tests, and tells you what happened. That shift from “AI tells me what to do” to “AI does the scoped thing I defined” is where the real productivity gains live. It requires you to get better at task definition: what are the inputs, what are the constraints, what does done look like. That is actually a skill most senior engineers already have. AI in the terminal is just finally making it worth exercising. The developers who will look like 10x engineers in 2026 are not necessarily the ones with the best AI prompts. They are the ones who understand their own workflows well enough to know where an AI agent can safely take over without supervision, and where a human needs to stay in the loop. Get that boundary right, and the terminal becomes a genuinely different place to work. Quick Start: The Minimum Viable Terminal AI Setup If you want to start today with just two things: Step 1: Install Claude Code. npm install -g @anthropic-ai/claude-code Navigate to your project directory and run claude. On first run it will ask you to authenticate with your Anthropic account. Step 2: Create your first CLAUDE.md. Spend 15 minutes writing down your project stack, your folder conventions, and three things you never want the AI to do (like using any in TypeScript or writing raw SQL). Put that file in your project root. That is the foundation. Everything else in this article builds on top of it. The claude.ai browser tab is not going away. But for the hours of your day that happen inside a terminal, the gap between developers who have integrated AI there and developers who have not is getting wider every month. The workflows above are not theoretical. They are running in production engineering environments right now, on real teams, shipping real code. The setup time for any of them is measured in minutes. The return starts the same day. Pick one. Set it up. See what changes. Follow me for more practical pieces on AI tools, developer productivity, and the workflows that are actually worth your time in 2026. A message from our Founder Hey, Sunil here. I wanted to take a moment to thank you for reading until the end and for being a part of this community. Did you know that our team run these publications as a volunteer effort to over 3.5m monthly readers? We don’t receive any funding, we do this to support the community. If you want to show some love, please take a moment to follow me on LinkedIn , TikTok , Instagram . You can also subscribe to our weekly newsletter . And before you go, don’t forget to clap and follow the writer️! 5 AI Terminal Workflows That Make You Look Like a 10x Engineer was originally published in Artificial Intelligence in Plain English on Medium, where people are continuing the conversation by highlighting and responding to this story.