Agentic Coding is AI’s Main Character; Here are 10 Tips to Get Started 🤖 💻

Claude Code - 10 Tips to Get Started ⚡

1. If you’re a visual person: run Claude Code in an IDE

Claude Code’s default interface is the terminal, fine if you love command lines, but a bit abstract if you crave visual cues.

What to do: Download an IDE (like Visual Studio, Cursor, or Windsurf) and integrate Claude Code so you can see what files are being edited, easily switch between different coding sessions, and highlight just the part of the code you want Claude to work on (this acts like your context‑window).

Bonus tip: You can make each session correspond to a branch or feature area (so you’re visually segmenting work) and use the editor’s built‑in diff/compare view to see what Claude changed.

2. Create a context.md file to outline the project

Think of this as your “what we’re building & how we roll” document. Create a Markdown file (e.g., context.md) at the root of your project. In it, outline the high‑level project requirements, constraints, style guidelines, team roles, etc. You can even ask Claude to generate a first draft based on your high‑level bullet points.

What’s the difference between .md and .txt?

• .txt is plain‑text, no formatting, just raw text.

• .md (Markdown) supports simple formatting (headers, bullet lists, code blocks, links). That makes it easy to read/edit and visually scan. Because it’s mostly text (no code), even non‑coders on your team can open/edit it.

Why it matters: context is everything for Claude Code, the better you define the project upfront, the more accurate (and less surprising) its output. Medium+1
Mini‑hack: Add sections like: “Project vision”, “Key features (MVP)”, “What won’t be included”, “Tech constraints”, “Style & naming conventions”, “Git branching strategy”. Then ask Claude: “Please refer to context.md before making any changes.”

3. Always start each section by asking Claude to refer to the context.md file

At the beginning of a session open VS/Claude Code and prompt: “Please read context.md and keep its contents in mind as you work.”

Why it matters: Starting fresh reminds Claude of the “big picture” before diving into details, this reduces drift, mis‑interpretation, and weird tangents.

4. Always double‑check its work… even if you don’t know how to code

Claude Code is powerful, but it’s not omniscient. After Claude makes changes, ask it:

• “Why did you do this change?”

• “Which file/line is this function implemented?”
  It will walk you through it. Over time you’ll pick up patterns (even if you’re not a full‑on coder).

Why it matters: Claude Code is like a very smart assistant who sometimes forgets what just happened. It can “drift” from your original instructions or act like it missed a step entirely. That’s why it’s key to keep checking its logic and asking it to explain itself.

5. Create “agent roles” that mimic SDLC team roles

Here’s where it gets fun: you’re not just using Claude, you’re organizing a mini‑team of agents. Define agents like:

• Product Manager

• Frontend Developer

• Backend Developer

• QA (Quality Assurance) Engineer

• DevOps Engineer

You can even add roles like “UX Designer”, “Data Engineer”, “Accessibility Specialist”.
Why it matters: This helps you frame prompts more precisely. Example: “Backend Developer agent: implement the API endpoint for user login per spec.” Your “QA agent” could then run tests. It’s a workflow separation that scales.
Mini‑hack: In your orchestration (see tip #6 in link below), label each agent and build hand‑offs: e.g., after “Frontend Developer” finishes UI, the “QA agent” checks browser compatibility, then “Security agent” reviews auth flows. That way it’s clear who did what.

6. Then create an orchestration.md file to specify when to call each agent

This is your workflow blueprint. Make orchestration.md at project root. Lay out which agent you trigger when, under what conditions. For example:

1. Product Manager drafts user story

2. Backend Developer builds endpoint

3. Frontend Developer consumes endpoint

4. QA Engineer runs tests

5. DevOps deploys to staging

Why it matters: Having this orchestration keeps your workflow clear (for you and Claude). It gives structure so Claude doesn’t randomly jump between roles or skip steps.
Mini‑hack: Use Markdown headings for each role, and under each heading list “Trigger conditions”, “Input”, “Output”, “Next step”. E.g., for QA Engineer:

• Trigger: Frontend shows “feature complete”

• Input: Branch feature‑xyz

• Output: Test report (pass/fail)

• Next: DevOps Engineer

7. Always sync your codebase to a cloud repo (e.g., GitHub)

Use a repo (GitHub or similar). Commit often. Push to remote. Tag milestones.

Why it matters: If Claude takes a wrong turn (or you decide you want to rollback), you’re covered. Also: version‑control helps you track “tech debt” or changes that need revisiting.
Mini‑hack: Ask Claude Code: “After you apply changes, commit with message ‘[agent‑role] completed feature X’.” That makes your commit history readable for humans (and future you).

8. Control tech‑debt: Claude seldom removes legacy code on its own

Claude is great at adding and optimizing, but not always great at cleaning up the old mess you don’t need anymore. Periodically ask: “Analyze the codebase. Which files/functions appear unused or redundant?” Then ask Claude to refactor or remove them explicitly.

Why it matters: If you neglect this, your codebase bloats, performance hits, readability drops… we call that tech debt.
Mini‑hack: Combine with your orchestration: after “feature complete” and “QA done”, add step: “Legacy cleanup agent runs a refactor pass.” Use a linter or code‑analysis tool (see tip #10) in that step.

9. Use commands (/…) regularly — the shortcut universe

Claude Code supports slash commands (and other CLI styled commands) to streamline workflows. Some useful commands:

• /compact — package up or collapse conversation context so you’re not wasting tokens

• /clear — clear conversation context to start fresh

• /install-github-app — integrate GitHub and have Claude review PRs

• /agents — list available sub‑agents or roles

Why it matters: Saves you from token waste, memory confusion, out‑of‑context prompts… basically, a neater workflow.

10. Use another coding‑agent (like OpenAI Codex) to complement it

You don’t need to put all your AI eggs in one basket. Let Claude do the initial heavy lifting and add another LLM to check against its work. Use OpenAI’s Codex (or equivalent) to run:

• Security vulnerability scans

• Code optimization & performance analysis

• Refactoring beyond syntax (algorithmic improvements)

Why it matters: Even though Claude is strong, a “second opinion” layer helps catch things that slip through: vulnerabilities, inefficiencies, architecture flaws.

Her AI Vault 🗂️

Your go-to stash of top resources we’ve created so you never have to dig through old emails to find that one gem.

📘 5 Ways To Get Started With AI At Work: Quick-start guide to delegate the “ugh” tasks and focus on what actually matters.

🔐 AI Privacy Guide: What your AI tools really collect and how to lock down your settings like a pro.

🔍 Her AI Glossary: 40+ essential AI terms, broken down in plain English.

🧠 AI Model Cheat Sheet: Claude, ChatGPT, Gemini... we decoded who does what best, so you can stop guessing and start impressing.

🛠️ Build-Your-Own Productivity Tool: A one-prompt web app for tasks, timers, and focus that is custom-built by Claude and copy-pasted by you.

Before You Go... 💭