Software Engineering

Delegating small coding tasks to LLM agents

Uses LLM chat and agent tools (e.g., CodeCompanion.nvim, Ollama/OpenWebUI, Crush with Copilot backend, Claude Code) to delegate tasks like implementing failing tests, debugging why code doesn’t work, and drafting larger changes that are written into the codebase.

Why the human is still essential here

The engineer scopes the work, provides context, evaluates the suggestions, integrates/edits code, and decides when/what to commit; responsibility for correctness and final changes remains with the human.

How people use this

Copilot-assisted unit test fix

Use an IDE assistant to propose a minimal patch that makes a failing unit test pass, then run tests locally and refine before committing.

GitHub Copilot (Copilot Chat) / VS Code

LLM-guided debugging from logs

Share a stack trace plus key source snippets to have the model suggest root cause hypotheses and a targeted fix you validate in your environment.

ChatGPT / Claude

Multi-file refactor drafted by an AI editor

Ask an AI-native code editor to perform a rename or API migration across multiple files and generate a coherent diff you can review and adjust.

Cursor

Community stories (1)

Blog

9 mins

How and why I attribute LLM-derived code

I’m a cautious skeptic of AI/LLMs, but I’m trying to use them where it makes sense in my software work. As I use chat/agent tools to help implement and debug code, I make a point of clearly documenting which commits (and sometimes lines) include LLM-derived code—often using Git commit trailers like `Co-authored-by` with the model name/provider—to improve traceability for reviewers and to reduce future legal/compliance risk.

Jamie TannaSoftware Engineer

Feb 25, 2026