You can spot someone who's good at Claude within about ten seconds of watching them work. They barely type. They skip the elaborate prompts, the over-explaining, the second-guessing. Just five specific moves that almost nobody teaches.
These are those five moves. If you've been using Claude for a while and feel like you've plateaued, the Claude hacks below are the gap between the outputs you're getting and the outputs you could be getting.
What you'll learn:
- Why briefs beat prompts
- How to use Claude Projects properly
- The interview-yourself technique
- Building custom tools with artifacts
- The screenshot-first workflow
1. Write briefs, not prompts
Most people treat Claude like a smarter Google. One question in, one answer out. Fine for trivia. Falls apart the moment you want something with taste attached.
The shift: stop writing prompts, start writing briefs. A brief gives Claude what any competent freelancer would ask for before starting work.
- Who the output is for
- What tone or format you want
- What you've already tried
- What a bad version would look like
That last one is the real unlock, and almost nobody does it. "Don't make it sound like a LinkedIn post" does more work than three paragraphs describing the tone you actually want. Claude is exceptionally good at steering away from things you name explicitly.
Prompt: "Write a product update email about our new search feature."
Brief: "Write a product update email about our new search feature. Audience: technical paying customers. Tone: direct, no marketing fluff. A bad version would start with 'We're excited to announce…' or use the word 'revolutionize.' Under 150 words."
Same ask. One gives you filler, the other gives you something you can actually send.
2. Use Claude Projects to stop repeating yourself
If you keep pasting the same context into every new chat (your codebase, your writing style, your brand voice, your product docs), you're doing it wrong. Claude Projects holds persistent context that every chat inside it inherits automatically. Set it once, benefit forever.
A few projects worth building today:
- Code review loaded with your team's style guide and past PR feedback
- Writing with brand voice, good examples, and the phrases you keep having to remove
- Support with product docs, FAQs, and tone guidelines
- Research that compounds across every question you ask
Ten minutes to set up, hours saved every week. The trick: treat the Project description like documentation for a new hire, not a throwaway prompt.
3. Let Claude interview you
This is the most underrated Claude hack on the list, and it looks backwards at first.
When you're stuck on something fuzzy (a spec, a strategic call, an architectural decision), don't ask Claude for the answer. Ask it to ask you questions.
"I'm trying to figure out [X]. Instead of giving me an answer, ask me five questions that will help me think about this more clearly. Start with the most important one."
What comes back is sharper than anything you would have written on your own. Claude surfaces constraints you hadn't named and assumptions you were making unconsciously. Answer the questions, then ask Claude to synthesize it all into the doc you actually needed. Your thinking. Claude's structure.
Build fast, scale faster
Backend infrastructure and web hosting built for developers who ship.
Start for free- Open source
- Support for over 13 SDKs
- Managed cloud solution
4. Build custom tools with artifacts in 30 seconds
Artifacts quietly replaces entire categories of software you used to search for. A JSON diff viewer, a regex tester, a timezone converter, a CSV to Markdown tool? Ask Claude to build it. You'll have it working in seconds, without wading through ad-filled search results or hitting signup walls.
The real power is iteration. "Add dark mode." "Copy to clipboard on click." "Remember my last five inputs." Each tweak is one sentence, and the tool shapes itself around how you work, not the generic assumptions of some random web app.
5. Screenshot first, describe later
Claude is multimodal and most people forget it. When you're debugging a UI issue, matching a design, or staring at a confusing error, stop describing and screenshot instead.
The visual context answers questions you wouldn't have known to ask. It works for broken layouts, chart interpretation, whiteboard photos, error panels, handwritten notes, and full PDFs. Anything you'd otherwise start explaining with "so basically it looks like…"
Bonus: annotate the screenshot before sending it. A red circle around the part you actually care about saves a whole round trip.
The common thread
Every one of these Claude hacks works for the same reason. They give Claude better context. Good context in, useful output out. Everything else is details.
The same principle applies to what you're building. Your app is only as good as the foundation under it, and if you're shipping fast, the last thing you want is to rebuild auth, databases, storage, and realtime sync from scratch every time.
That's the gap Appwrite fills. It's an open source backend platform that handles the parts of your stack that don't need to be custom, so you can focus on the parts that do. Auth with 30+ login methods, databases without managing infrastructure, storage, Functions, and Realtime subscriptions ready to use. Self hosting if you want full control.
Claude handles the thinking. Appwrite handles the platform. Pick one hack from above and try it this week, pair it with a few Claude Code tips and best practices for the terminal side of your workflow, then start your next project on Appwrite and skip the backend work entirely.
