Why Kiro Feels Like Cheating (But It's Not)

member_6e43b340 — Fri, 05 Dec 2025 14:16:09 +0000

I had an ambitious idea for Kiroween: a tool where “programmer ghosts” (AI Agents) analyze your legacy code through real-time voice. The stack was complex: React, TypeScript, Gemini Live API (Audio) and 6 different AI personalities.

Normally, this would take weeks. With Kiro, it took just a few days. Here’s how Context and Steering Docs became my secret weapon.

From Idea to Code: The Process

Kiro removed friction at every critical stage:

⚡ Instant Bootstrapping

I asked for a React + Vite + Tailwind project. Kiro generated the structure, settings, and dependencies without me touching the terminal.

👻 Character Generation

It created 6 “ghost” profiles (from Marcus the Hacker to Beatriz from DevOps) with their own prompts and voice styles in seconds.

🎧 Gemini Live Integration

The hardest part—processing PCM 16kHz audio—was solved by Kiro on the first try. It generated WebSocket logic, audio buffers, and handlers without me reading Google’s technical docs.

The Speed Secret: `.kiro/steering/`

This is where I won the race. I spent 15 minutes creating Steering Docs to give the IDE proper context. That changed everything.

I created files like:

.kiro/steering/horror-theme.md
.kiro/steering/tech-stack.md

With clear rules such as:

“Always use a sarcastic tone with references to death.”
“Color scheme: #0a0505 (black), #ef4444 (blood red).”
“Stack: Functional Components, TypeScript strict mode.”

The Result:

❌ Without Steering:

I had to constantly correct the AI:

“Use the correct red”, “Write it in TypeScript”, etc.

✅ With Steering:

Kiro generated perfect components on the first try, with correct styling and the sarcastic tone. It felt like working with a senior developer who had already read the entire documentation.

Autopilot and “Vibe Coding”

I enabled Autopilot for complex tasks like:

“Implement an Undo/Redo system (Ctrl+Z).”

Kiro created the hooks, history logic, and visual UI with zero supervision.

It also understood the project’s vibe: when I asked for a linter, it didn’t just detect errors—

it added looping scream audio whenever the code was broken.

It understood the horror theme perfectly.

Conclusion

Code Necromancer now includes:

Real-time voice chat
Code analysis
Flow visualization
A fully immersive horror atmosphere

The main lesson:

Kiro is not just autocomplete.

If you give it the right context (Steering Docs), it becomes a technical co-founder who knows your project as well as you do.

Tags: #kiroween #ai #react #gemini-api #productivity

DEV Community: member_6e43b340