Warp

The terminal hasn’t fundamentally changed since the 1970s. Think about that. Every other tool developers use has been reimagined multiple times. Editors. Browsers. Design tools. The terminal? Still a blinking cursor on a black screen, scrolling text into the void.

Warp fixes this. Built from scratch in Rust by Zach Lloyd, a former Google principal engineer, it’s a GPU-accelerated terminal that treats command output as structured data instead of a stream of characters, has AI built into every interaction, and behaves like software designed this decade. As of April 2026, the client is open source under MIT and AGPL v3, with 700,000+ developers and 37,000+ GitHub stars in the first week of public release.

I resisted switching. I’d been using the default macOS Terminal and occasionally iTerm2 for years. “It’s just a terminal,” I thought. Then I started using Claude Code daily, and the time I spent in the terminal jumped from 15 minutes a day to several hours. At that volume, every friction point compounds. The moment I realized I couldn’t go back was when I opened iTerm2 on a colleague’s machine and instinctively tried to click on a block that didn’t exist.

Why the terminal needs reinventing

If you’ve only used the default terminal, you might not realize how much friction you’ve been absorbing.

The output problem: every command’s output scrolls into a single continuous stream. Run five commands in a row and their outputs blur together. Want to copy the output of the third without grabbing pieces of the second and fourth? Good luck.

The editing problem: want to click in the middle of a long command to fix a typo? In most terminals, clicking selects text instead of placing a cursor. You’re pressing arrow keys repeatedly or memorizing shell key combos nobody remembers.

The knowledge problem: you know what you want to do but not the exact syntax. So you open a browser, search past three ads, find a Stack Overflow answer from 2014, adapt it, paste it back. A 15-second task took 3 minutes.

The collaboration problem: a teammate asks “what error did you get?” You screenshot the terminal or paste a wall of text into Slack that loses its formatting.

Blocks: the core innovation

Every command you run and its output are grouped into a discrete, selectable unit called a block. This sounds small. It isn’t.

You can click to select a single block — no more accidentally selecting output from adjacent commands. You can navigate between blocks with a keyboard shortcut instead of scrolling line by line. Looking for output from 50 commands ago? Jump to it in seconds.

You can share a block — click the link icon and Warp generates a shareable URL. Your teammate sees exactly the command and its output, properly formatted. You can bookmark blocks. When debugging, bookmark where the error first appeared, then always jump back without scrolling. Each block shows metadata: exit code, execution time, timestamp. Visible at a glance instead of buried.

For anyone running Claude Code sessions that generate dozens of command-output pairs, blocks transform the experience from scrolling a wall of text to navigating structured history.

AI features

Warp’s AI is accessed via a keyboard shortcut. Three modes solve real problems.

Natural language to command. Type what you want in plain English and Warp generates the exact terminal command. Instead of Googling syntax, you describe the goal, review the generated command, hit Enter, done. This is the feature I use most — not because I can’t write commands, but because the long tail of commands I use rarely would otherwise require a context switch to a browser. Find flags, awk patterns, complex regex, ffmpeg options. Warp keeps me in the terminal.

Error explanation. When a command fails, highlight the error and Warp analyzes it in context — not just the error message, but the command that produced it, your environment, and working directory. For opaque error messages from build tools and package managers, this saves real time.

Agent Mode. The most powerful feature. It runs multi-step plans directly in your shell, chaining commands, reading output, and self-correcting. Describe a goal — set up a Node.js project with TypeScript and a testing framework — and Warp plans the steps, runs each command, checks output, adjusts if something fails, and continues. I set up a complete CI/CD pipeline for a client using Agent Mode last month. Twenty minutes of checking documentation became four minutes of reading the plan and nodding.

The editing experience

This is where Warp feels like an IDE. Click anywhere in your input and the cursor goes there — sounds basic, but most terminals don’t do it. Multi-cursor editing with Option+Click. Selection and copy that work the way you’d expect. A command palette like VS Code’s. Persistent sessions that survive closing a tab. Split panes for running multiple processes side by side.

Pairing with coding agents

In 2026, the common pattern is running coding agents inside Warp with native support for Claude Code, Codex, Gemini CLI, and OpenCode. Warp provides the terminal, blocks, vertical tabs for parallel agents, navigation, notifications when agents finish, and built-in code review. Your coding agent provides the agentic loop: reading your codebase, writing code, running tests, managing git.

The block structure is especially valuable during agent sessions. When Claude runs a test suite and it fails, the failure output is in a clean block you can scroll to, copy, or share. The vertical tabs feature lets you run multiple agents in parallel — each gets its own tab with its own scroll history, and Warp sends a notification when an agent finishes. Start a Claude Code session on one branch, a Codex session on another, switch between them without losing context.

Warp Drive

A shared repository for your team’s terminal knowledge. Save common commands with descriptions and parameters — “Deploy to staging,” “Reset the test database,” “Generate a new migration.” Instead of these living in one person’s head or a dusty wiki, they’re accessible by name. Save parameterized templates where team members fill in variables and run.

We’ve been gradually converting our “only Illia knows how to do this” operations into parameterized workflows. The goal is that no deployment or database operation depends on a single person being available. We’re about 60% there.

Privacy: the honest conversation

I’m including this because it matters and most Warp guides skip it.

The code is now auditable — the client source is on GitHub. The community spent five years asking for this. But Warp still requires an account (your terminal activity is associated with an identity). Telemetry collects commands, output, exit codes, working directory, timestamps, and git branch by default. You can disable it in Settings. The open source release means you can verify that claim yourself.

AI features send terminal context to model providers (OpenAI, Anthropic, or Google depending on the model). Don’t run commands with API keys, passwords, or sensitive credentials while AI is active. This isn’t unique to Warp, but the terminal is where credentials flow most freely. I learned this mildly when Warp’s AI helpfully tried to autocomplete a database connection string that included a password. Nothing leaked, but the fact that the password was briefly in the AI context was enough to rethink my habits.

My approach: Warp for all general development work. For anything involving production credentials or sensitive infrastructure, I either disable AI features or switch to a plain terminal. If privacy is a non-negotiable hard line, Alacritty (GPU-accelerated, Rust-based, fully local) or iTerm2 (macOS, no account required) are solid choices. You lose AI and blocks, but everything stays local.

Pricing

The terminal itself is free forever. Paid tiers are about AI usage and team features. Free gets 75 AI credits per month (150 for the first two months). Build ($20/month) gets 1,500 credits plus bring-your-own API key. Business ($50/user/month) adds SSO, enforced Zero Data Retention, and shared team credits. If you already pay for an Anthropic API key, it works for Warp’s AI features.

The bottom line

Warp is the best terminal available in 2026 for people who use AI tools and coding agents. The block-based output, AI command generation, vertical tabs, and modern editing solve real friction that traditional terminals have ignored for decades. The open-sourcing resolved the biggest trust objection.

The remaining tradeoff is the account requirement. Your terminal activity is associated with an identity and AI context goes to external model providers. For most development work, this is a reasonable exchange. For sensitive infrastructure work, keep a secondary local terminal.

If you spend meaningful time in the terminal — especially running Claude Code or other agents — install Warp and use it for a week. The block navigation alone is worth the switch. The AI features are a bonus. And the editing experience will make you wonder why you tolerated a 1970s text input for so long.

---

This is the free web edition of Chapter 4. The full text — with installation walkthroughs, configuration files, shell integration scripts, Agent Mode session examples, and Warp Drive setup guides — is available in 42: The AI Builder’s Stack, coming Q3 2026 on Amazon in hardcover, paperback, and digital.