Anthropic just dropped a 31-page prompting guide.
Here's everything you actually need (in 10 rules):
1. You write "review this contract" and pray.
Fix: Name every output. "Review this contract. Flag risks per clause. Rate severity 1-5. Return as a table."
2: You say "summarize this" on a 40-page report.
Fix: 4.7 sizes the answer to the input. Cap it: "5 bullets. Each under 15 words. Start each with an action verb."
3: You write "don't use jargon. don't be salesy."
Fix: Negative instructions don't stick.
Flip them: "Write in plain English a 16-year-old could read aloud."
4: You type "can you help me with the email?"
Fix: Each verb ships something. For example: "Go to Gmail. Find [contact]. Write the send-ready reply. Under 90 words. Tone: confident, casual."
5: You wait for Claude to web search on its own.
Fix: Claude opus 4.7 calls fewer tools than 4.6.
Force it: "Use web search aggressively. Verify every claim with at least 2 sources."
6: You miss the warm tone from old Claude.
Fix: Claude opus 4.7 is direct. Almost zero emojis. Paste 2-3 sentences in the voice you want.
Tell Claude to match the rhythm.
7: You ask for "a landing page" & get bare minimum.
Fix: Drop this one line on every creative task
→ "Go beyond the basics."
It's from Anthropic's own doc.
8: You forget Claude 4.7 doesn't reason by default.
Fix: They call it "adaptive thinking."
Add this at the end: "Think before answering (maximum reasoning)." Free upgrade. Every time.
9: You rewrite the same prompt 14 times a week.
Fix: A skill is a command with instructions pre-built.
Write the same prompt twice? Make it a skill.
10: You assume Claude knows what you meant.
Fix: Old Claude 4.6 guessed.
New Claude 4.7 does exactly what you typed.
Spell it out. Output. Order. Length. Tone. Format.
If you don't say it, you don't get it.
To go even further & download my .md files directly:
Step 1. Go to https://t.co/psB7XxB2Y4.
Step 2. Subscribe for free. Don't pay anything.
Step 3. Open my welcome email (most skip this).
Step 4. Hit the automatic reply button inside.
Step 5. Download my .md files from my Notion.
Bonus. Enjoy my best copy-paste prompts, too.
@tangero Pokud bude na projektu pracovat více lidí, bude v .docs za chvíli pořádný guláš.
Druhák, o tom jak má vypadat CLAUDE.md, si vystačím s tím, co doporučují Andrej Karpathy nebo Boris Cherny.
@Suryanshti777 Good point, but greenfield needs different flow:
1) Interview - goals, arch, constraints → CLAUDE.md
2) Lock in patterns
3) Then run claude-code-setup
Without that, "analyze codebase" has nothing to work with.
@OpenAIDevs Cool feature. But which regions is this actually available in? Announcing things without mentioning availability is just wasting everyone's time. For most of the world this tweet is pure FOMO marketing, not useful information.
@dr_cintas Is it intentional that /codex:rescue supports inline --model, but /codex:review and /codex:adversarial-review only pick the model from config.toml and can’t override it per run?
@petarivanovv9@AntonMartyniuk Could you walk through the concrete order in which you’d add test scenarios (and which layers you mock or defer) when doing AI‑first TDD for something as simple as CreateUser in a brand‑new project?
𝗔𝗜 𝗮𝗴𝗲𝗻𝘁𝘀 𝘄𝗶𝗹𝗹 𝗿𝗲𝗽𝗹𝗮𝗰𝗲 𝗷𝘂𝗻𝗶𝗼𝗿 𝗮𝗻𝗱 𝗺𝗶𝗱𝗱𝗹𝗲 .𝗡𝗘𝗧 𝗱𝗲𝘃𝗲𝗹𝗼𝗽𝗲𝗿𝘀
Read this to survive 👇
Junior and mid-level devs often skip the fundamentals and take shortcuts on every path with AI.
AI amplifies what you already know.
If your fundamentals are weak, you can't tell when Claude or Copilot generates code that's subtly wrong.
And it will be wrong.
Often enough to matter in production.
The stronger your fundamentals, the better your judgment on what to keep, what to fix, and what to throw away.
Here are 5 fundamentals that will help you survive in the age of AI:
𝟭. 𝗛𝗧𝗧𝗣
→ Junior: ships the AI-generated controller.
→ Senior: catches the missing idempotency on POST, the wrong status codes, the leaked stack trace in error responses.
𝟮. 𝗖𝗼𝗻𝗰𝘂𝗿𝗿𝗲𝗻𝗰𝘆
→ Junior: ships the async method.
→ Senior: catches the missing CancellationToken passed through the call chain, and the .Result call that will deadlock under load.
𝟯. 𝗘𝗙 𝗖𝗼𝗿𝗲 𝗮𝗻𝗱 𝗱𝗮𝘁𝗮𝗯𝗮𝘀𝗲𝘀
→ Junior: sees code that compiles.
→ Senior: catches the N+1 query, the missing AsNoTracking, the IQueryable that materializes a million rows in memory.
𝟰. 𝗔𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲 𝘁𝗿𝗮𝗱𝗲-𝗼𝗳𝗳𝘀
→ Junior: accepts Microservices because Claude suggested it.
→ Senior: asks "why not start with a Modular Monolith and extract later if needed?"
𝟱. 𝗩𝗮𝗹𝗶𝗱𝗮𝘁𝗶𝗼𝗻 𝗮𝗻𝗱 𝗲𝗿𝗿𝗼𝗿 𝗵���𝗻𝗱𝗹𝗶𝗻𝗴
→ Junior: keeps try/catch inside the endpoint.
→ Senior: knows Result<T> beats exceptions for business flow, and FluentValidation beats inline if-else.
The AI didn't generate worse code in any of these cases.
The senior just knew what to look for.
That's the gap. And it's growing.
Developers who treat AI as an assistant they guide through structured decisions will ship production software.
Tomorrow, I'm sending my complete 10-step workflow to 𝟮𝟱,𝟬𝟬𝟬+ .NET developers:
✅ The CLAUDE[.]md template I use for every .NET project (the single most impactful technique)
✅ The 4 specification documents that turn AI into a Principal Developer
✅ The 3-pass code review prompt that catches N+1, missing auth, and leaked secrets
✅ The file reference trick that makes Claude follow your patterns instead of inventing new ones
This is the workflow I use to ship well-architected, production-ready .NET software with AI, without sacrificing quality. And you can too.
📌 Subscribe to my weekly .NET newsletter so you don't miss the issue:
↳ https://t.co/DCf4WsBaNV
——
♻️ Repost to help other .NET developers survive in the AI era
➕ Follow me ( @AntonMartyniuk ) to improve your .NET and Architecture Skills
@bcherny Great, will try it! Though with Opus 4.7 burn rate, I expect it to 1-shot book the outbound flight - and leave me with zero credits to book the return.
@karpathy Love this, but worth calling out the tradeoff explicitly: richer HTML UIs = more output tokens. We’re basically paying extra per answer to buy more “visual bandwidth” into the human brain - and that’s often a great deal.
@davidfowl Could you write a standalone article for senior devs new to AI-first: brutally practical from-scratch guide - Claude Code CLI/Desktop, CLAUDE.md, skills, plugins, MCP servers → multi-agent workflow with virtual AI collaborators across PRD → impl → test → review?
@rubenhassid Small nuance on the “negative rule”: Google’s official DESIGN.md spec has a canonical “Do’s and Don’ts” section where explicit “don’t” guardrails are expected. The real issue isn’t negatives, but negatives without a paired positive instruction.
ZipIt 1.4 just got a massive power upgrade ⚡️
Turn any website into:
📦 Full offline bundle (React / Vue / Next ready)
🎨 Design system (colors, gradients, fonts — copy instantly)
🧩 Export Clean components + Figma-ready layers (paste directly into Figma) , Download or collect in figma
🤖 AI prompts for ChatGPT / Claude / Gemini — recreate UI precisely , copy prompt to create any ui element
🖼 Extract all assets (WebP, Lottie, SVG, JPG, MP4, audio)
🧹 Production-ready clean CSS
♿ Full WCAG audit + downloadable PDF reports
All in one click.
Download → Inspect → Extract → Rebuild
https://t.co/QUI0Iqpt08 🚀