Software Engineering

Deciding when to delegate coding to AI vs. code manually

Apply a deliberate workflow (switching between different “modes” of AI assistance) to choose when autonomous agent work is worth it versus when direct manual edits are faster and clearer.

Why the human is still essential here

Human judgment is required to pick the right level of autonomy, provide accurate context, and ensure correctness, maintainability, and intent through oversight and iteration.

How people use this

Inline completion for tiny edits

Use inline suggestions for small, local changes (renames, one-liners, repetitive edits) instead of prompting an agent with a long explanation.

GitHub Copilot (inline completions) / JetBrains AI Assistant

Chat-based design exploration before coding

Ask an assistant to propose approaches, tradeoffs, and pseudocode for a feature, then implement the final solution manually once the shape is clear.

ChatGPT / Claude

Delegating a multi-file refactor to an agent

Hand off a well-scoped refactor (e.g., migrating an API client or updating deprecated calls across files) to an agent that opens a PR, then thoroughly review and test.

GitHub Copilot Workspace / Cursor (agent mode)

Community stories (1)

LinkedIn

Would you fly in a vibe coded airplane?

Would you fly in a vibe coded airplane?

That question has been rattling around my head as I've watched the AI coding discourse split into two camps: "AI writes all my code now" and "AI is overhyped garbage."


The reality is messier. After months of building real projects with AI agents on GitHub Copilot, I've found that the developers getting the most out of AI aren't the ones who went all-in on autonomy. They're the ones who learned when to use it and when not to.


I wrote up the framework I use — four modes of working with AI, from hand saw to chainsaw. The key insight: staying in one mode all the time will slow you down.


A single line of code is faster to change by hand than to describe to an agent. Some things are harder to explain than to code. And sometimes you don't know what you want until you start writing it yourself.


The inverse is equally true. Hand-typing boilerplate when an agent can generate it in seconds is a waste.


The skill isn't using AI. It's knowing when to use which mode.

CG
Cory GwinSoftware Engineer
Feb 23, 2026