Skip to content
BSWEN
Blog

Why Was GitHub Copilot Considered Terrible and Is It Actually Good Now?

Cowrie
Cowrie
Dev @ Bswen

Developer coding with AI assistance

Problem

I keep seeing developers dismiss GitHub Copilot as “terrible” in discussions. But then I see others saying it’s “an absolute powerhouse” now.

When I asked about this discrepancy on Reddit, I got clear answers: “Copilot used to be crap for a long time so people moved on and it’s pretty hard to rebuild a reputation.”

So I looked into what actually happened.

What Happened

The Early Days Problem

GitHub Copilot launched in 2021 as the first major AI coding assistant. It generated massive hype. But early versions had serious issues:

  1. Poor code suggestions - Often introduced bugs or security vulnerabilities
  2. Limited context understanding - Suggestions ignored project-specific patterns
  3. Agent mode launched late - Compared to Cursor/Claude Code, it was notably buggy on release
  4. Premium request bug - A critical bug counted each tool call as a premium request, making usage unpredictable
  5. Brand confusion - Microsoft’s “Copilot” branding across all products created association with the consumer AI chatbot’s shortcomings

From Reddit discussions:

  • “github copilot was terrible in the early days so it kinda has a bad reputation”
  • “Copilot agent launched relatively late and it was bad when it first launched”
  • “They rushed things, to be first to market, but it sucked. Hard to rebuild the reputation”

The Timeline

AI coding assistant quality timeline
| Year | GitHub Copilot       | Claude Code      | Cursor    |
|------|---------------------|------------------|-----------|
| 2021 | Launched (poor)     | Not available    | Not avail |
| 2022 | Improving (mediocre)| Claude 2 (limit) | Not avail |
| 2023 | Agent (buggy)       | Claude Code (good)| Launched |
| 2024 | Subagents (strong)  | Excellent        | Excellent |
| 2026 | "Absolute powerhouse"| Excellent        | Excellent |

What Changed

Copilot fixed its core problems:

  1. Subagents Architecture - Introduced specialized subagents for different coding tasks (planning, execution, review), matching Cursor and Claude Code capabilities

  2. Custom Agents - Allowed developers to create tailored agents for specific workflows and project patterns

  3. Premium Request Bug Fix - Corrected the tool call counting bug so each prompt counts as one premium request, not each tool invocation

  4. Improved Model Selection - Better integration of OpenAI’s latest models

  5. VS Code Integration - Deepened integration with native VS Code features

One Reddit user summarized: “Early Copilot honestly wasn’t that great to use. But since they added things like subagents and custom agents, plus the pay-per-request model, it’s become an absolute powerhouse.”

But another noted: “It’s good now, but Claude Code was good when GHCP was trash.”

The Reputation Lag Effect

This is the key issue.

Why the bad reputation persists:

  • Developers who tried early Copilot and found it inadequate switched to alternatives (Cursor, Claude Code, Zed)
  • These users established workflows around superior products
  • They have no incentive to switch back
  • Word-of-mouth from early negative experiences continues to influence new users
  • The AI coding assistant market is competitive - users evaluate products once and rarely revisit

The brand confusion problem:

Microsoft Copilot (consumer chatbot) quality issues create false negative perception for GitHub Copilot (coding assistant). They are completely different products, but the shared branding confuses people.

As one user noted: “Microsoft Copilot branding taints GitHub Copilot reputation.”

Common Mistakes People Make

  1. Assuming GitHub Copilot equals Microsoft Copilot - They are completely different products
  2. Judging current Copilot based on 2021-2023 reviews - The product has fundamentally changed
  3. Not testing subagent and custom agent features - These differentiate current Copilot
  4. Overlooking that premium request counting bug is fixed - Usage is now predictable
  5. Ignoring the speed advantage - Copilot is noted as “blazing fast” compared to alternatives

Current Configuration Optimization

If you’re using Copilot now, you can optimize it:

Example Copilot configuration
{
  "github.copilot.advanced": {
    "agents": {
      "planningAgent": {
        "model": "gpt-5-mini",
        "temperature": 0.3
      },
      "codingAgent": {
        "model": "gpt-5",
        "temperature": 0.1
      }
    },
    "useSmallModelFor": ["title-generation", "summaries", "simple-refactoring"]
  }
}

This configuration:

  • Uses smaller models for non-critical tasks
  • Optimizes premium request usage
  • Matches model capability to task complexity

My Recommendation

If you last tried Copilot before late 2024, test it again with:

  1. The new subagent features
  2. Custom agents for your workflow
  3. Configuration to use smaller models for simple tasks
  4. Experience the speed advantage current users highlight

The “blazing fast” response times combined with subagent architecture make it worth reconsidering.

The Key Lesson

First impressions in the AI coding market are disproportionately sticky. A product that launches poorly may permanently lose users even after achieving technical excellence.

This isn’t just about Copilot. It’s a pattern:

  1. Rush to market with imperfect product
  2. Users try it, find it inadequate
  3. Users switch to better alternatives
  4. Product improves significantly
  5. Original users never return
  6. Reputation persists despite improvements

GitHub Copilot earned its terrible reputation through a rushed, buggy early launch. The product has fundamentally improved. But the reputation damage persists because developers rarely revisit products they abandoned.

Summary

In this post, I explained why GitHub Copilot had a bad reputation and whether it’s worth using now. The key point is: Copilot improved significantly with subagents, custom agents, and bug fixes, but reputation damage persists because developers rarely revisit abandoned products.

If you dismissed Copilot based on early experiences, consider testing the current version. The subagent features and speed improvements make it competitive with Claude Code and Cursor today.

Final Words + More Resources

My intention with this article was to help others share my knowledge and experience. If you want to contact me, you can contact by email: Email me

Here are also the most important links from this article along with some further resources that will help you in this scope:

Oh, and if you found these resources useful, don’t forget to support me by starring the repo on GitHub!

Comments

Tags: