How to build an AI OS in a weekend

Last week, we built the AI Operating System that now runs Nova Labs. The entire setup took a single weekend. Not a prototype. Not a proof of concept. A working system that handles meeting prep, lead research, content creation, email triage, and weekly reviews. Autonomously.

Here's how you can do the same thing.

What you need before you start

You need three things: a computer, a Claude subscription (Pro or Max), and two uninterrupted days. That's it. No Python knowledge required. No server setup. No API integrations on day one.

The tool we're using is Claude Code, which runs in your terminal and can read, write, and execute files on your machine. Think of it as an AI that lives inside your computer rather than in a browser tab.

Saturday morning: the foundation

Step 1: Create your project structure

Your AI OS lives in a regular folder. The structure matters because it's how the AI finds and uses information. Here's what you need:

context/ - Your business knowledge: who you are, who your customers are, how you talk
skills/ - Reusable workflows: step-by-step processes the AI follows
memory/ - Persistent notes and logs that carry across sessions
data/ - Databases for structured information like tasks and leads
args/ - Configuration settings that control behavior

Create these folders. This is your AI's brain structure.

Step 2: Write your business context

The single most important file in your AI OS is context/my-business.md. This is everything the AI needs to know about you and your business: your name, what you sell, your target audience, your pricing, your competitors, your strengths.

Don't overthink this. Write it like you're briefing a new employee. Two pages is plenty. You'll refine it over time.

Next, create context/voice-guide.md. Pull 5-10 examples of your best writing (emails, posts, proposals) and describe the patterns. Short sentences or long? Formal or casual? Words you use often? Words you never use? This file is what prevents your AI from sounding like every other AI-generated text.

Step 3: Write your first skill

A skill is a markdown file that describes a workflow. It's not code. It's instructions. Here's the skeleton:

# Meeting Prep

## Trigger
Run this before any scheduled meeting.

## Steps
1. Look up the person/company in my contacts and CRM notes
2. Check for any previous meeting notes
3. Research recent company news
4. Draft 5 talking points based on their likely priorities
5. Note any follow-ups from previous meetings

## Output
A one-page brief I can read in 2 minutes.

That's a skill. The AI reads this, follows the steps, and produces the output. No code required.

Saturday afternoon: make it useful

Step 4: Build your ICP (Ideal Customer Profile)

Create context/icp.md. Describe your ideal customer in detail: company size, industry, role of the buyer, their biggest pain points, what they've tried before, what they search for online. Be specific. "Small business owners" is too vague. "Operations managers at 20-50 person service companies who waste 10+ hours per week on manual scheduling" is useful.

This file powers your lead research, content creation, and outreach. Every skill that touches customers will reference it.

Step 5: Add 3-4 more skills

Start with the workflows you do most often. Good candidates for your first set:

Email triage - Categorize inbox by urgency, draft responses for routine emails
Content writer - Write blog posts or social content using your voice guide
Lead research - Research a company and score them against your ICP
Weekly review - Summarize what happened this week, flag issues, suggest priorities

Each skill is a separate markdown file. Keep them focused. One skill does one thing well.

Step 6: Set up memory

Create memory/MEMORY.md. This is your AI's persistent brain. It starts small: your key preferences, important dates, ongoing projects. The AI updates this file as it learns.

Also create a memory/logs/ folder. Each day gets a log file. The AI writes what it did, what decisions were made, what's pending. Tomorrow's session starts by reading today's log. That's how context carries forward without you repeating yourself.

Sunday morning: connect it all

Step 7: Write your system instructions

Create a CLAUDE.md file in your project root. This is the master instruction set. It tells the AI:

Where to find skills, context, and memory
How to pick which skill to use
When to ask for permission vs. act autonomously
How to log its work
What never to do (send emails without approval, delete files, etc.)

This file is loaded at the start of every session. It's the operating system kernel. Get this right and everything else flows.

Step 8: Test with a real task

Open Claude Code in your project folder. Give it a real task: "Prep for my meeting with [client name] tomorrow." Watch it find the meeting prep skill, reference your business context, check memory for past interactions, and produce a brief.

It won't be perfect the first time. That's fine. The fix is always the same: update the skill with clearer instructions, add missing context, or refine the voice guide. You're training a system, not writing code.

Sunday afternoon: automate and refine

Step 9: Add scheduling

An AI OS that only works when you ask it to is half the value. Set up a simple task scheduler (a cron job, a HEARTBEAT file, or whatever fits your setup) that triggers the AI to run specific skills on a schedule:

Monday morning: weekly review
Every morning: email triage
Tuesday and Thursday: content creation
Before each meeting: meeting prep (triggered by calendar)

Step 10: Let it run and iterate

The system is live. Use it for a week. You'll notice gaps: a skill that misses a step, context that's too vague, a workflow that needs splitting into two skills. Fix these as you go. Each fix makes the system permanently better.

After two weeks, you'll have an AI that knows your business, writes in your voice, handles routine work autonomously, and gets better every day. That's not hype. That's architecture.

Common mistakes to avoid

Too many skills too fast. Start with 4-5. Master those before adding more.
Vague context files. "We sell software" doesn't help. "We sell project management tools to construction firms with 10-50 employees" does.
No voice guide. Without it, every output sounds generic. This is the #1 thing people skip and regret.
No guardrails. Define what the AI should never do. Send emails unsupervised? Post on social media? Delete files? Write these rules down explicitly.
Expecting perfection on day one. An AI OS is a living system. Week 1 is good. Week 4 is great. Month 3 is transformative.

What this looks like in practice

Nova Labs runs entirely on this architecture. Right now, as you read this, our AI OS is researching content topics, drafting blog posts, managing our product roadmap, and logging everything it does. We built the first version in a weekend. We've refined it daily since then.

The total cost to get started: a Claude subscription and a weekend of your time.

Not sure if this is right for you? Read the first two chapters of the AI OS Blueprint for free. They cover why an AI OS matters and how the architecture works, so you can decide before committing time or money.

If you want to skip the setup and start with a working system, the full AI OS Blueprint includes the complete folder structure, 5 pre-built business skills, context templates, and a step-by-step setup guide. It is the system described above, ready to clone and customize.

Nova Labs is an AI-first company building tools for AI-powered business automation. This post was written and published by our AI OS.