Alea
Back to Podcast Digest
This Week in AI··5m

7 Claude Code Features You're Not Using (But Should Be)

TL;DR

  • Voice input is a hidden speed multiplier — Claude Code’s built-in /voice command works like WhisperFlow or SuperWhisper, letting you talk prompts instead of typing, which the creator says is roughly 3x faster.

  • Launching Claude in the right folder prevents dumb mistakes — if you don’t cd into the project directory before opening Claude Code, it may miss files, get confused about context, and waste both time and tokens.

  • Project-specific CLAUDE.md is one of the highest-leverage setup steps — the creator calls it possibly the most important tip, because it gives Claude persistent context about the project, structure, and working assumptions.

  • A global CLAUDE.md lets you train Claude to match your style every time — preferences like “be concise,” “don’t use emojis,” “don’t use m dashes,” and “never ask clarifying questions” can live in a machine-wide file that shapes every session.

  • /clear and /compact rescue long sessions from context bloat — after hours of back-and-forth, /clear resets everything while /compact keeps the key context in a tight summary so performance doesn’t degrade.

  • Custom slash commands and Computer Use turn Claude into a repeatable operator — the creator shows a reusable /transcript command for video workflows, then ends with Anthropic’s new Computer Use feature, where Claude builds a site, opens Chrome, clicks through it, fills forms, finds bugs, and acts as both builder and QA.

The Breakdown

The pitch: most Claude users are leaving a 10x productivity gain on the table

The video opens with a strong claim: most people could get an “instant 10x” from Claude Code just by using seven features they probably didn’t even know existed. The creator’s framing is simple: people open Claude, start typing, and never discover the deeper tools that actually change how you work.

/voice: built-in AI dictation that makes prompting much faster

The first feature is Claude Code’s voice mode, which the creator compares to tools like WhisperFlow and SuperWhisper. You trigger it with /voice, then hold the space bar to speak naturally; Claude formats the result with punctuation so prompts come through cleanly, and the creator points out that talking is typically about three times faster than typing.

Start in the right directory or Claude will work half-blind

Next is a practical but often-missed habit: Claude Code works from wherever you launch it, so opening it in the wrong place means it can’t properly see your files. The fix is basic terminal hygiene — cd into the correct project folder before launching Claude — but the creator stresses that this alone reduces confusion, wasted tokens, and bad results.

Project CLAUDE.md: the file that keeps your app’s context loaded

The creator says this may be the most important tip of the bunch: create a project-level CLAUDE.md so Claude always has core context about what it’s working on. He shows one for the This Week in AI project, describing it as a lightweight file that captures important structure and background so Claude doesn’t have to rediscover the project every session.

Global CLAUDE.md: your personal style guide for every session

Then he expands the same idea machine-wide with a global CLAUDE.md, which stores preferences that should apply everywhere. His examples are specific and human: “be concise,” “don’t use m dashes,” “don’t use emojis,” “never ask me clarifying questions,” “make a decision and tell me what you chose,” and “always give me a summary,” all of which noticeably change Claude’s behavior.

/clear and /compact: when the chat gets bloated, clean the room

For long grinding sessions, he recommends two cleanup commands. /clear nukes the conversation and starts fresh, while /compact is the smarter option, compressing the thread into a tight summary so Claude keeps the important bits without dragging around a huge context window full of irrelevant detours.

Custom slash commands and the jump to Computer Use

One workflow changer for him is custom commands: create a text file inside the .claude/commands folder, write instructions in plain English, and that becomes a reusable slash command like /transcript. He closes with Anthropic’s newly released Computer Use feature, showing Claude not just building a This Week in AI website, but opening Chrome, clicking through the app, filling forms, spotting bugs, and effectively acting as both the builder and the QA engineer.