You’ll win
Valuable stuff today…
I've spent an unreasonable number of hours inside Claude Code. Here are the three lessons I’ve learned I wish someone told me on day one so that I could’ve saved hundreds of hours and be way more productive (like black and white).
First time hearing about Claude Code? Start with Anthropic's official quickstart first, then come back. Everything below will make more sense with that context. No, it’s not “only for devs“. If you have that belief, you’ll be outperformed by the marketer or designer that learned to use it and didn’t complain.
Lesson #1 (beginner friendly)
Context is the only resource that matters
Here's something nobody tells you when you start using Claude Code: the AI isn't limited by intelligence. It's limited by memory. Senior software engineers are using Opus 4.6. You are too. Yet they’re building 10x better stuff in 10x less time, so here’s how you can close that gap.
Claude Code has a "context window." Think of it as a whiteboard. Everything you say, every file it reads, every command it runs goes on that whiteboard. And the whiteboard has edges. When it fills up, Claude starts forgetting things. It'll contradict what it said ten minutes ago. It'll break patterns it just established. It gets sloppy.
Most people treat Claude Code like a group chat. They dump everything in and hope for the best. Power users treat it like a whiteboard in a strategy meeting. Write what matters. Erase what doesn't. Keep the surface clean.
Type /context , it’s a built in skill. I’m showing the Command Line below, but it works exactly the same in any other Claude Code interface (desktop, web, …)
You’ll get a detailed usage breakdown.
Just 24%, looks like we can keep going
Clear between tasks. The /clear command wipes the conversation and resets the whiteboard. If you're switching from debugging to building a new feature, clear first. The old context is noise now.
Compact when sessions get long. The /compact command tells Claude to summarize what's happened so far and keep working from the summary. Think of it as taking a photo of the whiteboard before erasing it. You lose some detail, but you keep the important decisions.
Use CLAUDE.md as permanent memory. This is a file you put in your project that Claude reads at the start of every session. Your coding conventions, your project structure, your preferences. It persists across sessions, so you're not re-explaining yourself every time you start fresh.
Watch the 70% line. Past 70% context usage, quality starts dropping. Past 85%, you'll notice real problems. Past 90%, stop what you're doing and clear. Pushing through a full context window is like trying to have a focused conversation in a room that's gotten way too loud.
If you take nothing else from this email, take this: context management is the single biggest differentiator between people who think Claude Code is "pretty good" and people who think it's the best tool they've ever used.
Lesson #2
Plan first, code second
The most common mistake I see (and I made it for weeks) is typing a request and letting Claude just... go. You describe a feature, hit enter, and Claude starts writing code immediately. AI slop after AI slop…
Sometimes it works, often it builds the wrong thing confidently, and you're three files deep before you realize it solved a problem you didn't have. If you’re a dev, it sucks. If you’re non-technical, it REALLY sucks, because you:
a) Want to fix everything on your own, and you end up burning thousands of tokens and hundreds of dollars, or
b) You go with it and end up putting live the king of AI slop, not only visually (looks like every other website) or something with serious security implications.
The fix is almost stupidly simple: make Claude think before it acts.
Claude Code has a Plan Mode (Shift+Tab twice, or just type /plan). In this mode, Claude reads your files, analyzes all the project, and proposes an approach. It can't edit anything. It just thinks out loud and waits for you to say "yes, that's right" or "no, here's what you're missing."
This is the workflow that changed everything for me:
Research. Tell Claude to read the relevant parts of your project. Don't let it start solving yet.
Plan. Ask for an implementation plan. "I want to add Google OAuth. What files need to change? What's the flow? Give me a plan." Claude will outline the steps, file by file.
Challenge. This is the part most people skip. Push back on the plan. "Why did you choose this approach over X?" or "What breaks if we do it this way?" I sometimes tell Claude to "grill me on these changes and don't write a single line until I pass your test." It forces both of you to think harder.
Execute. Only now does Claude start coding, with a clear map of what it's building and why.
But isn’t this this slower?
For the first ten minutes, yes. But the first time you watch Claude execute a clean plan start to finish with zero corrections, you'll never go back. This is THE HACK for non technical people. This is how the marketing guy is building an outreach engine or a custom CRM.
When to skip it: if you can describe the change in one sentence. "Fix the typo in the footer." "Make this header black." For anything that touches multiple files or involves a decision, plan first.
Lesson #3
You're a manager now, not a coder
This applies to everyone: engineers, content creators, marketers, designers… Forget everything you were taught about building stuff.
If you speak English and you have think in a structured way, you’ll win.
You're not the person writing code anymore. You're the person deciding what gets built, reviewing the output, and course-correcting when things drift. You went from individual contributor to engineering manager.
And if you've never coded before, the good news is you're starting with the right mindset already. You think in terms of "what do I need this to do" rather than "how do I implement this."
And you can manage Claude remotely now. Start a task from your terminal, then check in from your phone or browser. I've kicked off refactors, gone to get lunch, and come back to a clean PR. It's a weird feeling the first time. You get used to it fast.
Resources I keep coming back to
These are the five resources that helped me the most. Bookmark the ones that match where you are right now:
Anthropic's Official Best Practices (beginners)
Written by the team that built it. Context management, Plan Mode, CLAUDE.md, multi-session workflows. Start here.
skills.sh by Vercel (intermediate)
Open marketplace of installable skills. One command and Claude learns your framework conventions, design system, and deployment patterns.
claude-code-best-practice (intermediate)
84 best practices from Anthropic docs, Boris Cherny (creator of Claude Code), and power users. Clone-ready skills, hooks, and subagents.
awesome-claude-code (intermediate)
Curated catalog of everything the community has built on top of Claude Code. Skills, plugins, orchestrators, hooks. Updated regularly.
claude-code-ultimate-guide (advanced)
Six months of daily usage. 41 diagrams, 271-question quiz, security threat database, production templates. The deep dive.
What did you think of this email?
Just reply to let me know any specifics. This newsletter is built on what you actually need.
If this was useful, forward it to that friend. Yep, the one that just popped into your mind.
Stefano (@stefano_martell on X)
Co-Founder, The AI Leverage by Bestapps.ai