From Zero to Product
Building with AI
Real conversations showing how a non-engineer built a complete product
One person + one AI, 7 hours to a working automation tool
Background
The Problem
Spending hours every month on repetitive work
- Pulling data summaries from spreadsheets
- Filtering hundreds of updates for key changes
- Compiling monthly reports
- Notifying the right people
The Goal
Automate the whole process, from 3 hours to 5 minutes
My Background (not an engineer)
- Some Python, written a few small scripts
- Zero frontend experience
- Used Claude Code before, but never built a full project
Requirements to Architecture: Clarifying Needs
Me: I want to build an automation tool to handle my monthly reporting
Claude: Sure, can you tell me:
1. What specific repetitive tasks are involved?
2. Where does the data come from? (Excel? System export?)
3. How do you want to use the tool? (CLI? Web app?)
Claude doesn't jump straight to writing code — it asks questions to understand what you actually need
Tech Choices: Recommendations Based on Your Skills
Me: Let's use Python, I'm more comfortable with it
Claude: Got it. For Python, I'd suggest:
- FastAPI for the backend (Python web server)
- Tailwind CSS (no manual CSS needed)
- YAML for config files
What do you think?
Claude tailors recommendations to your skill level and asks for your input — it doesn't decide for you
It also explains the reasoning behind each choice, so you understand the architecture decisions
Core Development: Hitting a Wall
Me: Build the data analysis — pull directly from the system API
Claude: Sure, I'll need:
1. API connection settings
2. Your API key or token
3. The data schema
[writes a bunch of API integration code]
Me: Wait, I just realized I don't have API access...
Company requires an approval process and it takes forever
Wrote a bunch of code only to discover it won't work — always verify your access and resources before building
After the Wall: Quick Pivot
Claude: Got it. Let's try a different approach:
Option A: Request API access from IT
Option B: Export CSV manually, process with code
Option C: Use a different API (requires different permissions)
Which is most feasible for you?
Me: B — no waiting on IT, I can start right away
Claude: OK, switching to CSV import.
[removes API code, writes CSV parser instead]
Wrong turns are fine — state your constraints and Claude pivots instantly. You make the decisions, Claude provides the options
Web UI: Even Without Frontend Skills
Me: I want a web interface, but I have zero frontend experience
Claude: No problem. I'd suggest:
- Tailwind CSS (use classes instead of writing CSS)
- Alpine.js (lightweight JS, feels like HTML attributes)
This way you barely need to touch JavaScript
but can still build a polished interface.
Me: Can the UI look nicer? Not too "engineer-y"
Claude: Sure, let me add rounded cards, gradients, animations...
How does this look?
You can give subjective feedback ("it looks weird", "too engineer-y") and Claude will understand and adjust
Notifications: Complex vs. Simple
Original Plan
- Apply for a developer account
- Set up a webhook server
- Manage tokens and channels
- 2-3 days of setup work
Alternative
- Create a bot (2 minutes)
- Get the token
- One HTTP POST to send messages
Me: That's too complex. Is there a simpler way to send notifications?
Claude: There's a simpler option... Your call?
Me: Let's go simple — getting it working is what matters
Voice your pain points ("too complex"), Claude offers alternatives. Ship first, optimize later
From Hardcoded to Team-Ready
Me: Right now every config change requires editing code
and redeploying. My teammates can't do that — only I can
Claude: Got it. You need an admin interface.
I'd suggest storing config in a database
so your team can manage it through the UI
Me: But this data is sensitive — what if someone messes it up?
Claude: Good point. I'll add safeguards:
1. Each user gets their own API key
2. Keys can be revoked or rotated anytime
3. All operations are logged
Raise your concerns, Claude adds protections. Explain the severity and it responds with appropriate caution
Pitfalls We Hit
Config files vanished after deploy
Config that existed locally was missing in the cloud environment
Moved sensitive config to environment variables
Database query limitations
Complex queries needed pre-configured indexes
For small datasets: fetch everything, filter in code
Don't panic when errors happen — paste the error to Claude and it'll handle it
Effective Collaboration Patterns
Give clear context
Your background, constraints, and why you're building this
Raise concerns
Security issues, performance worries, things you're unsure about
Give real feedback
Not just "ok" or "not ok" — explain why something works or doesn't
Ask it to explain
"Why did you do it this way?" "Are there other approaches?"
You make the decisions
Claude provides options, you choose. It's the engineer, you're the product owner
Take small steps
One feature at a time. Build, test, then move to the next one
Time Stats
Total dev time
Learning + building alone
Time saved
One non-engineer, one AI, 7 hours to a working automation tool
The most valuable thing isn't the code AI writes,
but the decisions made through the conversation.
From Zero to Product: Building with AI — End
Use arrow keys to navigate