This post is both a portfolio case and a product creation story. Writing it as it happened, no sugarcoating.
The Idea
I was working with AI agents and kept hitting the same problem: the agent gives an answer, but why it arrived at that answer is unclear.
For simple tasks, that’s fine. For complex ones — it’s a disaster. If the agent made a mistake at step 3 of 10, you only find out from the final result. By then you’ve already spent tokens, time, and nerves.
First Prototype
Started simple: visualizing reasoning traces as a tree. Each model call — a node. Each branching — a fork.
Sounds simple, but the devil is in the details:
- How to parse streaming traces?
- How to display a tree with 200 nodes?
- How to make it interactive?
How I Built It
I’m not a programmer in the traditional sense. I orchestrate AI:
- Define the architecture and specification
- Decompose into tasks
- AI writes code for each task
- I review, test, iterate
This isn’t “press a button and get an app.” It’s managing AI like a junior developer who types very fast but needs a clear spec.
Takeaways
- Speed: from idea to MVP in 2 weeks (one person)
- Quality: 80% of what a senior developer would write
- Bottleneck: not code, but product decisions — what to build and why
That’s exactly why I believe the future belongs to people who understand both product and technology. You don’t need to be an expert in both — you need to be good enough to orchestrate.