Anthropic Claude Code

Claude Code is Anthropic's attempt to turn its models into an actual coworker rather than a chat window. Instead of pasting snippets back and forth, you hand it a task - a bug fix, a missing test suite, a migration that might run for days - and then steer and review its work from wherever you happen to be: the terminal, VS Code or JetBrains, the web app, a phone, GitHub, or a Slack thread. The headline capability is agentic autonomy: reading and searching a codebase, spawning sub-agents, running commands, and editing across multiple files to land a change.

The surface coverage here is genuinely unusual. Terminal-first tools rarely also ship desktop builds for macOS, Linux and Windows, mobile apps, and a Slack integration, and MCP support means it can be wired into whatever internal tooling and data sources a team already runs. That breadth signals who this is really for: professional developers and platform teams who already live in a shell, and who want an agent that can be supervised from a phone while a long job grinds away. Developers who want a polished, point-and-click assistant welded into a GUI editor are not the target audience, and shouldn't expect this to feel like one.

Commercially, it's bundled rather than sold - there's no separate licence and no free tier. Pro at $20/month (or $17 annualised, paid $200 up front) is the way in; Max at $100 and $200/month buys 5x and 20x the usage plus early feature access and priority during busy periods. Teams get a $100/seat Premium tier with a five-seat floor, Enterprise adds a larger context window, SSO and compliance tooling at custom pricing, and there's a pay-per-token API route for anyone who'd rather meter it.

On the evidence available, this is a strong and unusually well-distributed product, and the $20 entry point is aggressive for what is nominally a full agentic coding tool. It's also a product whose real cost is hard to predict. Usage on paid plans is governed by a five-hour rolling window on top of a weekly cap, and nothing in the material we have quantifies those limits in tokens, requests or hours. For a tool explicitly pitched at multi-day migrations, that's the single most important number for a buyer, and it isn't published - which makes "from $20" a starting price rather than a realistic one. The five-fold jump from Pro to the $100 Max tier strongly implies that serious sustained use lands well above the headline figure.

Two other gaps are worth flagging. The model lineup is described as Sonnet, Opus and Haiku with the top tier restricted or metered on cheaper plans, but that detail rests on a single third-party plan guide rather than Anthropic's own documentation, so treat any assumption about which model you actually get on Pro with caution. And the only performance signal we have is one external editorial score of 9.0/10 praising autonomous multi-file work while conceding the CLI-centric workflow is more demanding than a GUI editor - encouraging, corroborating the design intent, but a single second-hand data point, not benchmark evidence. We have no independent measurements of code quality, failure rates, or how well the agent behaves when a long-running task goes wrong.

So: a confident recommendation on capability and integration breadth, a heavily qualified one on value. Existing Pro or Max subscribers should simply use it - it's already paid for. Teams evaluating it as a purchase should budget for a Max or Team seat and treat the five-hour/weekly cap structure as the thing to test in a trial before committing. Four stars, with the missing usage-limit and quality data the reason it isn't higher.