Welcome to The AI Brief.

Every week, you get clear, honest breakdown of what’s really happening in AI. No recycled press releases pretending to be insight. Just the developments that actually matter.

The AI industry spent the last year obsessing over model releases, benchmark scores, and which company is winning the race. Almost nobody asked the more practical question: are people actually using these tools well?

The answer, in most cases, is no.

Claude Code has over 50 built-in slash commands sitting idle in the terminals of developers who have never pressed the forward slash to find them. Those commands handle code review, security scanning, CI debugging, task scheduling, and autonomous execution, replacing tools that cost most teams $150 to $200 a month in separate subscriptions.

This issue covers what they do, which ones cost extra credits, and where the genuine limits are.

Let's get into it.

That polished proposal? Dictated in two minutes. The detailed follow-up email? Spoken between meetings. The Slack update with clean formatting and exact numbers? Voice.

Wispr Flow turns everything you say into professional, send-ready text. Speak naturally — go on tangents, change your mind, circle back — and get output that reads like you spent twenty minutes writing it.

89% of messages sent with zero edits. Millions of professionals use Flow daily. Works in every app on Mac, Windows, and iPhone.

Start flowing free

There is a version of Claude Code that most developers know: describe what you want, it writes some code, you review it and move on. Useful, certainly, but roughly equivalent to having a fast pair programmer you have to supervise at every step.

Then there is the version most developers do not know exists, and it works completely differently. It reviews pull requests autonomously, runs security audits on your pending changes, executes tasks in parallel across multiple worktrees, and runs scheduled routines on Anthropic's cloud whether your terminal is open or not. You do not supervise this version. You set it up, and it runs.

The gap between those two versions is a forward slash.

Type / in Claude Code's terminal and a menu appears with over 50 built-in commands. Most developers using the tool have never seen this screen, not because it is hidden, but because Anthropic has done a genuinely poor job of surfacing it, and most people simply do not think to press slash.

Here is what some of those commands actually do.

/plan forces Claude into pure architecture mode before it writes a single line of code. No edits, no files created, just the strategy laid out for your review. Run this before any major change and you stop Claude from charging into the wrong implementation and spending 30 minutes undoing it, which is the thing that quietly wastes the most tokens.

/batch breaks a large task into independent parallel units, each running in its own git worktree with its own subagent. Adding error handling across 30 API endpoints becomes a single command. The caveat matters: if your task units are not genuinely independent, you will get merge conflicts. This is a tool for repetitive isolated changes, not interconnected refactors, and knowing the difference is what makes it useful.

/schedule is where Claude Code stops feeling like a tool and starts feeling like infrastructure. It creates routines that run on Anthropic's cloud, not just while your terminal is open, so you can set up morning PR summaries, automated fixes when CI fails, or code reviews that trigger without you initiating anything. A tool responds when you talk to it; infrastructure runs regardless.

/compact addresses a problem most developers feel without being able to name it. Long sessions degrade — Claude loses the thread, contradicts earlier decisions, forgets context it established an hour ago. /compact compresses the full conversation into a focused summary and resets the quality of the session without losing the work you have done. Run it every 30 to 40 minutes during any heavy session.

What these commands collectively represent is a version of Claude Code that Anthropic built and, for reasons that are genuinely unclear, chose not to make obvious.

The capability to replace project management tools, code review platforms, security scanners, and CI debuggers has been sitting idle behind a single forward slash that most paying users have never pressed. That is not a minor discoverability issue. That is the whole story.

The /loop command runs Claude on repeat until a task is done. A developer set it to fix 47 failing tests, stepped out for coffee, and came back to green tests, a zero credit balance, and a charge matching a mid-level contractor's day rate. Three hours of agent calls, no ceiling set, no one watching. The tests passed. That is something, at least.

Right. Now for the part where we actually take a position.

Claude Code, used properly, is not a coding assistant in any conventional sense. It is closer to a programmable development environment that can plan, review, execute, schedule, and monitor its own output without constant supervision. Calling it a coding assistant is like calling a database a spreadsheet — technically adjacent, functionally a different thing.

The honest frustration is that this capability exists, works, and is largely invisible because of how the product was introduced to most users. New users open Claude Code, type a prompt, get a response, and conclude they have seen what the tool can do. They have not seen it. They have seen the lobby.

The developers getting real value from Claude Code are the ones who pressed the forward slash and kept reading the menu. There are not nearly enough of them, and the responsibility for that sits with the company more than the users.

This issue was about what is already in your terminal, unused. Next time you open Claude Code, type the forward slash before you type anything else and spend two minutes reading what appears.

If you want the full reference PDF of every command covered here, with usage notes and the warnings most write-ups leave out, reply to this email. I will send it directly.

And if something here changed how you see Claude Code, or if you think I got something wrong, reply and tell me. Both are equally welcome.