Claude Code
Anthropic's agentic coding tool, run from a terminal, an IDE, a desktop app, or a browser, with the same project memory and MCP servers on every surface.
Who it's for
Developers who already live in a terminal or a real IDE and want an agent that plans, edits across files, and runs commands itself, on the same project memory whether they're at their desk or on their phone.
Standout details
- One engine runs everywhere: terminal CLI, VS Code and JetBrains extensions, a desktop app, claude.ai/code in the browser, and iOS/Android, all reading the same CLAUDE.md and settings
- Subagents let a lead agent split a task and run pieces in parallel, and the Agent SDK exposes the same tools to build fully custom agents
- Routines run scheduled or event-triggered tasks (a nightly dependency audit, a PR review on every push) in the cloud, so they fire even when your machine is off
- MCP support plus hooks and shareable skills mean it can read a Jira ticket or a Google Doc and turn a slash command into a packaged team workflow
- Usage is metered against your plan's session and weekly allowance, not sold as a flat seat, and that allowance has moved under users more than once: an unannounced tightening in July 2025, a second weekly cap on Opus usage added that August, and a bug in March 2026 that drained quotas far faster than normal across every paid tier before Anthropic fixed it
What it is
Claude Code is Anthropic’s own coding agent: it reads a codebase, plans a change, edits across files, and runs shell commands and git itself. It started as a terminal tool and now runs on the same engine from a VS Code or JetBrains extension, a standalone desktop app, a browser at claude.ai/code, or the Claude mobile app, with CLAUDE.md project instructions, auto-built memory, and connected MCP servers carrying over between all of them.
Beyond single-session edits, it covers the surrounding workflow: subagents and the Agent SDK for building custom orchestration, hooks for running shell commands around its own actions, skills for packaging a repeatable workflow behind a slash command, and Routines for scheduling work (a PR review, a dependency audit) to run in the cloud on a timer or a GitHub event.
Who should use it
Developers who are comfortable in a terminal and want an agent that does more than autocomplete: multi-file refactors, its own test-and-fix loop, and commits and PRs it opens itself. It fits builders who work across several surfaces in a day, since the same project context follows a session from a desk to a phone through Remote Control. It sits next to Cursor in this category; Cursor bakes the agent into a VS Code fork, while Claude Code stays CLI-native first and adds the editor as one surface among several.
It’s a weaker fit for someone who wants a single flat-fee, no-terminal experience, or who writes only a handful of lines a day and would never touch subagents, hooks, or Routines.
What stands out
The multi-surface consistency is real: the same CLAUDE.md, settings, and MCP connections apply whether the session started in a terminal or a phone browser, and claude --teleport or the /desktop handoff move a running session between them. Subagents and background agents let one task fan out into several running at once, coordinated by a lead agent, which is a different shape of work than a single-threaded chat-style assistant.
The usage-limit history is worth knowing before you buy a plan. Anthropic quietly tightened Claude Code’s session limits in July 2025, then added a second weekly cap specific to Opus usage that August, changes developers first noticed as complaints rather than an announcement. In March 2026, a bug briefly caused far faster quota drain than normal across every paid tier, including Max 20x accounts, before Anthropic acknowledged and fixed it. None of that makes the tool worse at coding; it does mean the allowance behind your monthly price is not as fixed as the sticker suggests.
The verdict
Claude Code earns its place for anyone whose daily work already happens in or near a terminal and wants an agent that can plan, edit, test, and ship without narrating every step. Pro at $17 to $20/mo is a reasonable entry point, but the real product is Max 5x and Max 20x, for people who run it most of the day.
The catch is predictability. You are buying a metered session and weekly allowance rather than unlimited seat access, and that allowance has changed shape more than once with little warning. Anyone who needs to know exactly what a month will cost, or who wants everything in a GUI with no CLI step at all, should look elsewhere first.
Honest verdict
For someone who already works in a terminal, Claude Code is the most complete version of this idea: one agent, the same memory and MCP servers, follows you from CLI to IDE to phone. The tradeoff is that you are buying a metered allowance, not a fixed seat, and Anthropic has changed what that allowance means more than once without much warning. If a predictable flat bill matters more to you than the deepest agent on the market, or you only touch code a few times a week, the $17 to $20/mo entry price buys more capability than you'll use. Skip it if you want a GUI-first, no-terminal experience end to end.
Published 2 Sept 2026