Let me be completely honest: I didn't write a single line of code for OpenOlex. Not today, not yesterday. The AI (Kiro) wrote 98-100% of it.
So what did I actually do? And why am I even sharing this?
What I Actually Did
1. Clarified What OpenOlex Is
The project kept drifting. "AI controls everything" → "Cross-device orchestration" → back to the original idea.
I spent time defining: OpenOlex is the MCP protocol's physical world interface. Not a robot. Not an app. Just infrastructure—a bridge between MCP and IoT devices.
I spent time defining: OpenOlex is the MCP protocol's physical world interface. Not a robot. Not an app. Just infrastructure—a bridge between MCP and IoT devices. bridge between MCP and IoT devices.
This clarity matters because it determines what gets built.
2. Decided: Tests Over Features
Day 1 had 45% test coverage. I could have told the AI to build more features. Instead, I said: "Write comprehensive tests for MQTT Broker and Gateway."
Result: 56% coverage, found a data loss bug, cleaned up 51 ESLint warnings.
The AI wrote all the tests. I made the decision that tests were more important than features today.
3. Reviewed What Got Built
I read through the test code. I checked the coverage reports. I verified the architecture still makes sense.
I didn't write it, but I'm responsible for it.
The Uncomfortable Truth
98-100% of the code is AI-generated. Here's what that means:
What the AI does:
- Writes all TypeScript code
- Implements all tests
- Fixes all bugs
- Handles all the "coding"
What I do:
- Identify gaps in the industry (MCP needs physical world interface)
- Define the project vision and positioning
- Make priority decisions (features vs quality)
- Review and validate what gets built
- Make architectural choices
Is this "real" software development? I don't know. But it's fast, and the code quality is actually good.
Why Share This?
Because I think this is going to be normal soon, and we should talk about it honestly.
Most "I built X with AI" posts hide how much AI did. They make it sound like the human wrote most of the code and AI just "helped."
That's not what's happening here. The AI is doing the heavy lifting. I'm doing something else—something closer to product management + architecture.
The question I'm exploring: If AI can write all the code, what's the human's actual value?
My current answer: Decisions and direction. The AI can implement anything, but it can't decide what's worth building or why.
Day 2 Numbers
- Test coverage: 45% → 56%
- ESLint warnings: 51 → 0
- Lines of code I wrote: 0
- Lines of code AI wrote: ~2000
- Time spent: Defining direction, reviewing output, making decisions
What's Next
Day 3: Storage abstraction layer, session management, push toward 60% coverage.
The AI will write it. I'll decide what and why.
About This Experiment
I'm building OpenOlex in public over 21 days using AI tools. It's open source. It's 100% AI-generated code.
Why? To see what happens when you let AI do all the coding and focus entirely on decisions and direction.
Follow along:
Day 2 of 21: Zero code written. All decisions made. 🤖
Real question: Is this still "software development"? Or is it something new? I genuinely don't know, but I'm documenting the journey. 👇
Top comments (0)