Daily engineering
ChatGPT/Codex and Gemini for implementation, review, debugging and fast technical cross-checks.
AI / 05
I use AI, but I do not hand production over blindly.
AI is an engineering tool for implementation, review, debugging, documentation and automation. The goal is not autonomous magic; it is faster work with controlled context and verifiable output.
WORKFLOW
Where I use it now
Local models
Ollama-based models for tasks where repository and technical context can remain local.
Repository knowledge
Documentation, source code and workflows connected through retrieval/RAG-style context rather than giant memory prompts.
Role-based agents
Engineer, documentation and Git agents with separate responsibilities, later extending into low-risk content and configuration automation.
GUARDRAILS
Control matters more than autonomy
- PR and quality gates before AI-modified code is accepted
- Tests and smoke checks instead of “looks plausible”
- Narrow tool permissions and responsibility boundaries
- Human decision for high-risk production actions