I got access to Claude Code and decided to test its capabilities by building a voice assistant superior to Siri on an old computer.
Wait, what claude code again?
Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster through natural language commands.
I’d been using Cursor’s AI tools, but Claude Code offered more autonomy, more “do the thing” energy.
The Challenge
Even small open-source models already outperform Siri by a wide margin. I built the project on a decade-old PC with:
- 16GB RAM
- NVIDIA GTX 970 (4GB VRAM)
- Ubuntu OS
- Technologies: Ollama, Gemma models, Whisper.cpp, and lightweight TTS
I chose C++ over Python, wanting to stretch its legs a bit rather than use the expected language.
What Followed Surprised Me
Claude Code generated complete project structure including source files, installation scripts, README documentation, and build processes. When errors occurred, the system self-corrected after receiving compiler feedback.
It Worked
The voice assistant functioned fully offline on the local machine using only open-source models.
Working With Claude Code
I developed a structured process: Define task → Build → Test → Self-review → Fix → Commit → Document.
Strengths:
- Full engineering support beyond code generation
- Documentation, scaffolding, packaging, and explanation
- Clean Git commits with quality messages
- Proper naming conventions and project structure
Weaknesses:
- Narrow focus on immediate tasks (viewing “the world through a toilet roll tube”)
- Struggled with high-level integration logic
- Needed constant guidance for complex streaming audio scenarios
- Unit testing capability exceeded integration testing ability
Impact and Reflection
The conversational nature of speaking to the AI felt different than text-based interaction, even using a 90s-style Stephen Hawking speech synthesizer.
This is what I do as a developer. Watching systems perform faster and cleaner triggers both intimidation and excitement.
The Age of Magic
LLMs will transform software development similarly to how compilers changed machine code approaches. Eventually, code may evolve into a new kind of machine language, running on new kinds of machines, built by systems that we won’t fully understand.
The age of magic is near. And if you ask me — it’s already begun.