Why I Stopped Using Agent Mode in GitHub Copilot
About this blog
I stopped using Agent mode in Copilot a few weeks ago. Hear me out why.
The Journey: From Stack Overflow to Agent Mode
Like everyone else, I started using Copilots a few months ago. Before that, I was just asking questions in ChatGPT, getting answers, and trying to validate them myself. It still felt like a replacement for Stack Overflow. I was thinking, I’m still in control.
But after I started using Agent mode, I gave Copilot full control. I became more like a manager or product owner - just telling it what I needed. As a developer, I do know how to approach problems, so I would explain the approach, and it would do everything else. Creating new files, modifying code, deleting things - everything.
Over that period, I felt like I became dumb. (Maybe I was already dumber.) But I genuinely felt it. I didn’t get the satisfaction that I had built something myself, even though productivity increased.
That feeling matters.
The Problem with Full Autonomy
When you use Agent mode in GitHub Copilot, you’re essentially handing over the wheel. The agent autonomously:
- Plans and implements changes across multiple files
- Runs terminal commands (though it asks for permission first)
- Invokes tools and makes decisions
- Creates, modifies, and deletes files
- Can commit and push code to repositories
For many developers, this feels like magic. Your productivity skyrockets. Tasks that would take hours are done in minutes. But there’s a hidden cost - your involvement in the actual craft of coding diminishes.
Even though the agent asks for permission before running terminal commands, the overall workflow encourages you to keep approving and moving forward. You become more of a reviewer than a creator, approving changes rather than crafting them yourself.
As developers, we don’t just write code to solve problems. We write code to understand problems deeply, to think through edge cases, to learn new patterns, and to build muscle memory. When an agent does all of that for you, you’re optimizing for output, not for learning or satisfaction.
Understanding GitHub Copilot Modes
Before we talk about how to switch modes, let’s understand what’s available in VS Code.
Three options in GitHub Copilot
GitHub Copilot in VS Code provides three different options:
-
Agent: Autonomously plans and implements changes across files, runs terminal commands, and invokes tools. This is the “full control” mode that I was using.
-
Plan: Creates a structured, step-by-step implementation plan before writing any code. Hands the plan off to an implementation agent when it looks right. This is a middle ground.
-
Ask: Answers questions about coding concepts, your codebase, or VS Code itself without making file changes. This is the mode I switched to.
How to Completely Disable Agent Mode (If You Want)
If you want to take it a step further and completely disable the Agent mode option from your VS Code, you can do that by editing your settings.json file.
- Press
Ctrl+Shift+P(Windows/Linux) orCmd+Shift+P(Mac) to open the Command Palette - Type “Preferences: Open User Settings (JSON)” and select it
- Add the following line to your settings:
|
|
This will completely remove the Agent option from your agent picker dropdown. You’ll only see Ask and Edit modes available.
Note: Your organization might also control this setting. If you don’t see the agent option or can’t change this setting, contact your admin to enable or disable this functionality.
My New Workflow: Minimal and Intentional
So the bottom line is: I started using it minimally. You can’t avoid it completely if you want to keep pace with how fast things are moving. But at the same time, you have to protect your well-being, your thinking ability, and your sense of ownership.
How I Use Copilot Now
- Ask mode for questions: When I need to understand a concept, debug an issue, or get unblocked, I use Ask mode. It gives me answers without taking control.
- Inline suggestions for boilerplate: I still use Copilot’s inline code suggestions for repetitive boilerplate code. It saves time without removing me from the coding process.
- Agent mode sparingly: For well-defined tasks where I’ve already thought through the solution and just need execution speed, I occasionally use Agent mode. But it’s the exception, not the rule.
The Environmental Cost
Also, we should be mindful of the environmental cost of using these models unnecessarily. Every AI request consumes computational resources and energy. When we use Agent mode for trivial tasks or let it run unchecked, we’re not just losing our own agency - we’re also contributing to unnecessary carbon emissions.
Using AI tools responsibly means being intentional about when and how we use them.
This Is Not Advice
This is not advice. This is just for developers who felt the same way I did.
If you’re someone who thrives on productivity and loves delegating repetitive work to AI, Agent mode might be perfect for you. But if you’re like me - someone who values the craft, the learning, and the satisfaction of building something with your own hands - then Ask mode might be the better fit.
The key is to find a balance that works for you. Use AI as a tool to enhance your capabilities, not replace them.
Summary
In this blog, we covered:
- Why I stopped using Agent mode in GitHub Copilot
- The difference between Agent, Plan, and Ask modes
- How to completely disable Agent mode using VS Code settings
- My new minimal and intentional workflow
- The environmental cost of using AI tools unnecessarily
At the end of the day, we’re still developers. We should be the ones in control, not the AI.
References
If you find this post helpful, please consider sponsoring.
