DEV Community: btechnik

Get Started with Claude Code: A Developer's First Look at AI-Powered Coding

btechnik — Fri, 03 Apr 2026 22:48:25 +0000

Introduction

There are a lot of AI tools out there. Claude.ai, GitHub Copilot, Cursor, ChatGPT — the list keeps growing. So when I sat down to explore Claude Code, I had one question in mind: what actually makes this different?

By the end of this project, I had a working portfolio website — built almost entirely by AI — and a clearer picture of where Claude Code shines. Here's exactly what I learned.

What Is Claude Code?

Claude Code is Anthropic's agentic coding tool that runs directly in your terminal. Unlike Claude.ai (which is a chat interface), Claude Code has real capabilities that go much further:

✅ Reads your entire project folder and understands file structure
✅ Creates, edits, and deletes files on your behalf
✅ Runs shell commands and terminal operations
✅ Remembers project context across sessions via CLAUDE.md
✅ Asks for permission before making changes (in Default mode)

The key insight: Claude.ai only knows what you tell it. Claude Code knows what's in your project.

Step 1: Installing Claude Code

Installation on Windows is a single PowerShell command:

irm https://claude.ai/install.ps1 | iex

Once installed, verify it's working with:

claude --version
# Output: 2.1.79 (Claude Code)

You'll also need to add API credits to your Anthropic account — Claude Code uses the API, so usage is token-based rather than subscription-based.

Pro Tip: Start with a small credit top-up ($5–$10) to explore the tool before committing more.

Step 2: Understanding the Trust Prompt

When you first run claude in a terminal, it immediately asks a critical security question:

"Is this a project you created or one you trust? Claude Code will be able to read, edit, and execute files here."

This isn't just a formality — it's an important safety mechanism. Claude Code needs read/write permissions on whatever directory you point it at. Running it from your home directory would give it access to personal files, OS configs, and sensitive data.

Best practice: Always navigate to a dedicated project folder before launching Claude Code. This scopes its permissions and keeps it focused on your actual work.

Step 3: Setting Up Project Memory with CLAUDE.md

This is one of the most powerful features, and easy to overlook.

Claude Code reads a file called CLAUDE.md at the start of every session. Think of it as a briefing document — it tells Claude Code what your project is, what tech stack you're using, conventions to follow, and anything else it should keep in mind.

# CLAUDE.md

This file provides guidance to Claude Code when working with code in this repository.

This is a single-page personal portfolio website built with plain HTML and CSS. No frameworks.

With this in place, Claude Code never asks what kind of project you're working on. It already knows.

Why this matters for teams: CLAUDE.md can include coding standards, naming conventions, deployment notes, and architectural decisions. It becomes a living knowledge base that any team member — human or AI — can rely on.

Step 4: Configuring Claude Code

Two key configuration commands: /model and /config.

Choosing a Model

I selected Claude Haiku — the fastest and most cost-effective option. For a portfolio website, the heavy reasoning power of Opus wasn't needed.

Key Config Options

Setting	My Choice	Why
Thinking Mode	Disabled	Not needed for HTML/CSS; saves tokens
Permission Mode	Default	Requires approval for every file change
Model	claude-haiku-4-5	Fast and cost-effective

Permission Modes explained:

Default — Claude Code proposes changes and asks for approval before applying
Auto-Accept Edits — Changes applied automatically without prompting
Plan — Claude Code only plans; you execute manually

For beginners, Default mode is highly recommended. It keeps you in control and helps you understand what the tool is doing.

Step 5: Building the Portfolio

With an empty project folder and a single prompt, Claude Code generated a complete portfolio website:

> Build me a portfolio website with About, Projects, and Contact sections. 
  Use a modern, technical color scheme.

Claude Code then:

Assessed the empty directory
Created index.html with semantic HTML structure
Added inline CSS with a dark, technical theme
Populated the About section with a professional bio layout
Added placeholder cards in the Projects section
Built a Contact section with social links

Total cost for the initial build: $0.01. Tokens: ~390 input, ~5.8k output.

Iterating in Plain English

After the initial build, refinements were done in natural language:

"Add my GitHub and LinkedIn links to the footer"
"Update the About section with these professional details..."
"Make the color scheme darker and more technical-looking"

Each time, Claude Code proposed specific changes, showed the diff, and waited for approval.

Step 6: The @ Reference System

The most impressive feature: Claude Code's @ folder reference system.

With markdown files for the three past projects in a /projects folder, I used this prompt:

> Update the Projects section in @index.html using the content from @projects/ folder. 
  Read each markdown file to extract the title, description, and tech stack.

Claude Code read all markdown files, extracted titles, descriptions, and tags, then updated the HTML with real project cards automatically.

The result: a fully populated Projects section featuring actual work — Telegram AI Bot with OpenClaw, Spotify Clone with Cursor, AWS Threat Detection with GuardDuty — all pulled from documentation files.

Monitoring Usage: /cost and /context

Two commands worth knowing:

/cost — Real-time session breakdown:

Total cost: $0.0593
Total duration (API): 34s
Total code changes: 655 lines added, 0 lines removed

/context — Context window usage:

31.4k tokens (16%) out of 200k tokens

Keeping an eye on /context matters for longer sessions — as the context fills up, responses about earlier parts of your project can become less precise.

Key Takeaways

It's not just a chat interface with file access. The combination of project awareness, CLAUDE.md memory, permission-based editing, and @ references creates a genuinely different developer experience.

The cost model is worth understanding. For the work covered here — building an entire portfolio website — total spend was under $0.10. Choosing the right model and disabling Thinking Mode when unnecessary makes a real difference.

Default permission mode is your friend. Seeing every proposed change before it's applied is an excellent way to understand what Claude Code is doing and why — which makes you a better developer in the process.

What's Next?

Next goals: deploying this portfolio via GitHub Pages, and exploring Claude Code with more complex projects — connecting it to AWS services, running test suites, and using Thinking Mode for actual architecture decisions.

If you've tried Claude Code, I'd love to hear about your use cases in the comments.

Nikhil Bhan is an AWS Community Builder (AI Engineering) based in Vancouver, Canada. He creates technical content for the NextWork community at nextwork.org.

I Built a Personal AI Assistant with a Telegram Bot Using OpenClaw — Here's How

btechnik — Fri, 03 Apr 2026 22:28:04 +0000

Author: Nikhil Bhan

🚀 What I Built

Imagine having an AI personal assistant that you can text from anywhere — your phone, your desktop, wherever — and it knows you, remembers context, and can even send you messages on a schedule without you asking.

That's exactly what I built using OpenClaw, a free and open-source AI agent framework. In under two hours, I had a fully functional, secured, autonomous AI assistant connected to Telegram and powered by Anthropic's Claude Sonnet 4.5.

In this post I'll walk you through exactly how I did it — installation, configuration, Telegram integration, security hardening, and cron-based automation.

🧰 What You'll Need

A Windows computer (PowerShell required for install)
An Anthropic API key (get one at console.anthropic.com)
A Telegram account (mobile or desktop)
~90 minutes of your time

🦀 What is OpenClaw?

OpenClaw is a free, open-source AI agent that runs locally on your machine. It acts as a gateway between you and AI providers like Anthropic. Key features:

Control UI — a browser-based chat interface to interact with your assistant
Gateway — a background process that keeps your AI available at all times
Channel integrations — connect Telegram, WhatsApp, Discord, and more
Cron jobs — schedule autonomous AI tasks to run on a timer
Plugins & Skills — extend what your assistant can do
Allowlists — restrict who can access your bot

It's the closest thing I've found to a personal AI operating system.

Step 1: Installing OpenClaw and Running the Gateway

Open PowerShell on your Windows machine and run:

powershell
iwr -useb https://openclaw.ai/install.ps1 | iex

This launches the OpenClaw Onboarding Wizard. During setup, you'll:

Choose your AI provider — select Anthropic
Choose your model — I selected Claude Sonnet 4.5
Enter your Anthropic API key

The wizard also installs the OpenClaw Gateway, which is a background service that keeps your AI assistant running even when you're not actively using the Control UI.

Once installed, verify everything is working:

openclaw gateway status

Troubleshooting tip: If you can't access the Control UI in your browser, check the PowerShell terminal running the Gateway for the correct IP address. I initially tried the wrong IP and spent time debugging — the correct one is shown in the terminal output.

Step 2: Your First Conversation in the Control UI

Open your browser and navigate to the IP address shown in the Gateway terminal (typically http://127.0.0.1:18789).

This is the OpenClaw Control UI, a clean chat interface where you can talk directly to your AI assistant.

Try these prompts to get started:

Hello! Can you introduce yourself and tell me what you can do?

Your assistant will list its capabilities — task automation, web search, calendar management, connecting to messaging platforms, running commands, and more.

Check token usage with /status:

/status

This gives you real-time session data: which AI provider is active, which model you're using, and how many tokens have been sent and received. It's excellent for monitoring API costs.

Step 3: Connecting Telegram for Mobile Access

This is where things get exciting. You can now message your AI from your phone.

3.1 Create a Telegram Bot

Open Telegram and search for @botfather
Send the command /newbot
Give your bot a name (I used MyNextWorkClaw)
Give it a username (I used teknikassistantbot)
Copy the bot token that BotFather provides

3.2 Add the Bot to OpenClaw

Go to your OpenClaw Control UI and send this message to your assistant:

Add a Telegram channel using this bot token: [YOUR_BOT_TOKEN]

Your assistant will update the openclaw.json configuration file automatically.

3.3 Pair Your Telegram Account

Open Telegram and search for your bot by username
Send /start — a pairing code will be returned
Go back to the Control UI and send:

Approve this pairing code: [PAIRING_CODE]

That's it. Your Telegram account is now paired with your AI assistant. Send it a message from Telegram — it will respond!

Step 4: Securing the Bot with an Allowlist

By default, anyone who finds your bot could message it and use your Anthropic API credits. We need to lock it down.

Get Your Telegram User ID

There are several bots that can tell you your User ID (such as @userinfobot). You can also see it in the pairing confirmation.

Update the Allowlist

In the Control UI, tell your assistant:

Update the OpenClaw configuration so that only Telegram User ID [YOUR_USER_ID] is allowed to use this bot.

Verify the change:

Show me the current OpenClaw configuration.

You should see your User ID in the allowedUsers field. Anyone else who messages the bot will be blocked.

Step 5: Automating with Cron Jobs

This is the feature that makes OpenClaw feel genuinely agentic. Instead of you always initiating the conversation, your AI assistant can proactively send you messages.

In the Control UI:

Set up a cron job that runs every 10 minutes and sends me a Telegram message with a motivational quote, a fun fact, and a random piece of trivia.

Your assistant will create the cron configuration. After 10 minutes, you'll receive a Telegram message from your bot — completely unprompted — containing exactly what you asked for.

This is a simple example, but the possibilities are significant:

Daily summaries of your calendar
Reminders for tasks or medications
Weather alerts
Web-scraped news digests

📊 What I Learned

Concept	What It Does
`/status` command	Tracks token usage and session info
OpenClaw Gateway	Keeps your AI running in the background
Bot pairing code	Securely connects your Telegram identity to OpenClaw
`openclaw.json`	The config file that controls everything
Allowlist	Restricts bot access to approved User IDs only
Cron jobs	Enables proactive, autonomous AI behavior

🔮 What's Next

This project opened my eyes to what a local AI agent can really do. My next goals with OpenClaw:

Build custom OpenClaw Skills — domain-specific capabilities for my workflows
Create automated AWS monitoring alerts — integrating with my GuardDuty log analyzer project
Connect to calendar and task tools — true personal assistant automation

💡 Why OpenClaw Over Just Using Claude.ai?

That's a fair question. Claude.ai is excellent for conversation, but OpenClaw gives you:

Persistent background availability — your assistant is always on
Channel flexibility — use Telegram, WhatsApp, Discord — not just a web tab
Cron automation — AI-initiated actions, not just reactive responses
Local control — your config, your API key, your data flow
Skills and plugins — extend behavior programmatically

🏁 Conclusion

In about 90 minutes, I went from zero to a secured, mobile-accessible, autonomous AI personal assistant. The combination of OpenClaw's simplicity, Anthropic's powerful Claude models, and Telegram's ubiquity makes this genuinely practical for everyday use.

If you're building in the AWS/AI/automation space like I am, I'd strongly encourage you to try this. The cron job capability alone changes how you think about what AI can do for you.

GitHub / Resources:

OpenClaw: https://openclaw.ai
Anthropic API: https://console.anthropic.com
NextWork Projects: https://nextwork.org

Nikhil Bhan is an AWS Community Builder (AI Engineering) based in Vancouver, Canada, with 13 years of experience in healthcare IT, systems administration, and technical documentation.