Can I really code from my phone?

Yes, but you're not typing code on your phone. You manage an AI agent (Kiro CLI) that codes for you via WhatsApp or Telegram. You act as Product Manager giving high-level architectural instructions. The agent handles implementation, testing, and Git commits. You review specs and approve execution from anywhere.

How much does this setup cost?

Infrastructure costs $0/month using AWS Free Tier (t4g.small with 2 vCPUs, 2GB RAM, 30GB storage, 750 hours/month). LLM costs: Free ChatGPT (limited) or ChatGPT Plus $20/month (unlimited) or Claude API $20-40/month. Total: $0-20/month depending on your AI choice.

OpenClaw (formerly Clawdbot/Moltbot) is an open-source personal AI assistant by Peter Steinberger and the community. You run it on your own machine/server, connect it to the chat apps you already use (WhatsApp, Telegram, Slack, Discord, iMessage, etc.), and then drive real work through skills and automations. It’s configured via an onboarding wizard (`openclaw onboard`) and supports injected prompt files like AGENTS.md, SOUL.md, and TOOLS.md to shape behavior and tooling.

Why use Kiro instead of Cursor or Copilot?

Kiro is spec-driven and designed for autonomous server-based operation. Unlike Cursor (requires an IDE open) or Copilot (autocomplete-focused), Kiro runs on servers 24/7, generates architectural specifications before coding, and handles multi-day tasks without human intervention. It can orchestrate changes across multiple repositories, making it ideal for mobile-first workflows.

Is WhatsApp safe for this workflow?

Treat any chat channel as untrusted input. WhatsApp automations often rely on unofficial libraries and can be fragile (and may violate platform ToS), so use a dedicated number and expect occasional breakage. Telegram is usually the safer “production” choice thanks to an official Bot API and predictable behavior. With OpenClaw, keep DM access locked down (pairing/allowlists), don’t let random senders trigger actions, and follow the OpenClaw security guide—misconfigurations can expose API keys and private data.

What are the limitations of a t4g.small instance?

The t4g.small has 2GB RAM which is insufficient for Node.js agents without swap memory. You must configure 4GB+ swap or builds will crash with OOM errors. Heavy builds (large Next.js apps with many dependencies) will be slow due to swap thrashing. For production workloads, upgrade to t4g.medium (4GB RAM) or larger.

Can I use this for production deployments?

No. This setup is for development and prototyping only. Production deployments require proper CI/CD pipelines (GitHub Actions, AWS CodePipeline) with automated security scanning (SAST/DAST), comprehensive testing, code review, and monitoring. The mobile workflow is for feature development, not production releases.

What happens if my AWS instance crashes?

Set up automatic daily backups to S3 (free tier: 5GB storage) using cron jobs. Install OpenClaw as a daemon (via `openclaw onboard --install-daemon`) so it survives reboots, and use tmux/screen for other long-running terminal work. Configure CloudWatch alarms for instance health monitoring (CPU, memory, disk). Keep all repos synced to GitHub so code is never lost. Instance state can be restored from S3 backups within 15 minutes.

What is swap memory and why is it critical?

Swap memory is disk space used as virtual RAM when physical RAM is exhausted. On a 2GB RAM instance, Node.js processes (especially npm install) will crash with Out of Memory (OOM) errors without swap. Configure 4GB swap minimum using fallocate. Swap prevents crashes but is slower than RAM, so expect performance degradation during heavy builds.

How do I prevent my AWS free tier from expiring?

AWS t4g.small free tier provides 750 hours/month (enough for 24/7 operation) for 12 months from account creation. After 12 months, you'll be charged $0.0168/hour (~$12/month). To continue free: create a new AWS account, migrate your setup, or switch to Oracle Cloud (always-free Ampere A1 with 4 vCPU/24GB RAM).

Can I use this workflow with other AI models?

Yes. OpenClaw and Kiro support any LLM with API access (OpenAI, Anthropic, Gemini, and more). Configure your provider during `openclaw onboard`. For cost optimization, use a fast "daily driver" model for 90% of tasks and reserve premium models for debugging and architecture reviews.

How I Built a $0 AI Coding Agency with OpenClaw (formerly Clawdbot)

Yesterday, I shipped 12 features while waiting for my kids’ soccer practice to end. Not by frantically typing on my phone, but by managing an AI development team that works 24/7. Here’s how you can build your own mobile coding agency for exactly $0.

Update (Feb 24, 2026): Clawdbot has been renamed to OpenClaw. This post now uses the new name (and updates the install commands).

The Setup That Changed Everything

What I built: A free AWS instance running OpenClaw (formerly Clawdbot) that turns WhatsApp messages into working code through Kiro CLI’s spec-driven development.

Total cost: $0-20/month (AWS free tier + optional ChatGPT Plus)
Setup time: 90 minutes
Result: I manage AI agents from anywhere, they handle the implementation.

TL;DR: My $0 AI Coding Agency

Cost: $0-20/month total (AWS free tier + free ChatGPT or $20 ChatGPT Plus)

Stack: AWS t4g.small + OpenClaw + Kiro CLI

Workflow: WhatsApp message → Spec discussion → AI builds → You review

Result: Ship 3-5 features weekly from your phone

Setup: 90 minutes total

Reality: You become product manager, AI becomes developer

Step 1: Claim Your Free AWS Computer

First, get your free cloud computer:

# AWS Console → EC2 → Launch Instance
# Choose: Ubuntu 22.04, t4g.small (ARM Graviton), 30GB storage
# Security Group: SSH (port 22) from your IP only

# Connect and add swap (critical for 2GB RAM)
ssh -i your-key.pem ubuntu@your-instance-ip
sudo fallocate -l 4G /swapfile
sudo chmod 600 /swapfile && sudo mkswap /swapfile
sudo swapon /swapfile
echo '/swapfile none swap sw 0 0' | sudo tee -a /etc/fstab

Step 2: Install Your AI Assistant

Now install OpenClaw (formerly Clawdbot) - think of it as your personal AI project manager:

# Install Node.js 22+ (required for OpenClaw)
curl -fsSL https://deb.nodesource.com/setup_22.x | sudo -E bash -
sudo apt-get install -y nodejs

# Install OpenClaw
npm install -g openclaw@latest

# Interactive setup wizard (recommended: installs a user daemon so it stays running)
openclaw onboard --install-daemon

During onboarding, you’ll connect:

LLM Provider: Free ChatGPT, ChatGPT Plus ($20/mo), Claude, or your preferred AI
Messaging: WhatsApp (via QR code) or Telegram (recommended for stability)
Personality & tools: Configure behavior via SOUL.md and tool access via TOOLS.md

Step 3: Add Your Coding Agent

Install Kiro CLI - the AI that actually writes code:

npm install -g @kiro/cli
gh auth login  # Connect GitHub
kiro init      # Enable spec mode (this is crucial)

The Magic: How It Actually Works

Here’s my real workflow from last week:

Me (from WhatsApp): “Build a Next.js app that tracks my workout progress with charts”

OpenClaw: “I’ll help you build that. Let me discuss the requirements first:

TypeScript or JavaScript?
Which chart library? Recharts or Chart.js?
Need user authentication?
Where to store workout data?”

Me: “TypeScript, Recharts, no auth for now, localStorage is fine”

OpenClaw: “Perfect. Here’s the spec I’m sending to Kiro:”

# Workout Tracker Spec

- Next.js 14 with TypeScript
- Tailwind for styling
- Recharts for progress visualization
- Local storage for data persistence
- Components: Dashboard, AddWorkout, ProgressChart
- Pages: Home, History, Analytics

Me: “Looks good, proceed”

Then Kiro takes over, working through the spec:

✅ Initializes Next.js project
✅ Sets up TypeScript and Tailwind
✅ Creates workout data models
✅ Builds dashboard with charts
✅ Commits each step to GitHub

I get notifications: “Task 3 complete: Dashboard built. Moving to charts…”

Total time: 18 minutes while I drank coffee.
Code quality? Solid. I reviewed it later and deployed.

Real Examples From My Phone

During commute: “Add dark mode to my portfolio site”
While grocery shopping: “Create API endpoint for user preferences”
At gym: “Refactor the authentication logic in my Express app”

Each time, the same pattern: Discuss → Spec → Approve → Build → Review

Why This Beats Traditional Development

Before: Sit at desk, open IDE, write code, test, commit, push
Now: Send message, review spec, approve, get notifications

The shift: You become the product manager, AI becomes the developer.

Security Setup (Don’t Skip This)

# Install fail2ban (blocks IPs after 3 failed SSH attempts - prevents brute force attacks)
sudo apt-get install fail2ban

# Disable password authentication (force SSH key-only access)
sudo nano /etc/ssh/sshd_config
# Set: PasswordAuthentication no
# Set: PubkeyAuthentication yes
sudo systemctl restart sshd

# Protect API keys (never commit these to GitHub)
nano ~/.env
# Add: OPENAI_API_KEY=your-key-here
echo 'export $(cat ~/.env | xargs)' >> ~/.bashrc

My Daily Workflow

Morning: Check overnight progress, plan new features
Lunch: Approve specs, start new builds
Evening: Review completed work, deploy to production

Result: 3-5 features shipped weekly without being tied to my desk.

What Works Best

✅ Perfect for: CRUD apps, dashboards, API endpoints, documentation
✅ Great for: Learning new frameworks, prototyping ideas. For framework comparisons to guide your choice, see our React vs Next.js vs Astro guide.
✅ Ideal for: Side projects, client work, MVPs

❌ Avoid for: Security-critical code, complex algorithms, production deployments without review

The Reality Check

This isn’t magic. You need to:

Understand the code Kiro generates
Review every spec before approval
Know when to say “no, try different approach”
Handle debugging when things break

But for 80% of development tasks? It’s transformative.

Understanding the code AI generates is critical. Our React performance optimization guide and TypeScript patterns guide will help you review AI output effectively.

Your 90-Minute Challenge

Ready to build your mobile development agency?

Launch AWS instance (20 min)
Install OpenClaw + connect messaging (25 min)
Install Kiro + connect GitHub (20 min)
Security hardening (15 min)
Test with “hello world” (10 min)

Tonight: Message your new AI team: “Build a todo app in React”

Tomorrow: Review the spec during your commute, approve it, and watch your phone light up with progress updates.

The future of development isn’t typing faster—it’s managing AI teams that never sleep.

Cost Optimization and Scaling Beyond Free Tier

As your AI coding workflow matures, here’s how to optimize costs and scale up:

Free Tier Alternatives

Provider	Free Offering	Best For
AWS t4g.small	750 hrs/mo for 12 months	Getting started
Oracle Cloud	Always-free Ampere A1 (4 vCPU, 24GB RAM)	Long-term free option
Google Cloud	e2-micro always free	Lightweight workloads
Hetzner	€3.79/mo (cheapest paid)	Post-free-tier budget option

When to Upgrade

Signs you’ve outgrown the free tier:

Build times exceed 10 minutes due to swap thrashing
Multiple repositories need simultaneous agent access
Team members want their own agent instances
Production previews require more compute

Multi-Agent Architecture

For larger projects, run multiple specialized agents:

# Terminal 1: Frontend agent
tmux new-session -s frontend
cd ~/projects/my-app && kiro start --scope frontend

# Terminal 2: Backend agent
tmux new-session -s backend
cd ~/projects/my-api && kiro start --scope backend

# Terminal 3: OpenClaw orchestrator
tmux new-session -s openclaw
openclaw start

Coordinate via OpenClaw: “Frontend agent: build the dashboard. Backend agent: create the API endpoints it needs.”

Want to learn more about AI-assisted development? Check out our Complete Guide to AI-Augmented Development comparing Cursor, Windsurf, Kiro, and more.

Concerned about code quality? Read Why 73% of Vibe-Coded Apps Fail to learn how to ship AI-generated code safely.