GitHub Copilot: My Developer Report After Four Years of Daily Use

In short: GitHub Copilot is an AI code completion and generation service by GitHub and Microsoft, integrated into VSCode, JetBrains, and other editors. Since 2024 it supports OpenAI, Claude, and Gemini models. Business and Enterprise tiers keep code private and GDPR-compatible, but every suggestion still needs human code review.

I've been using GitHub Copilot daily since late 2021 — I was in the preview program before the public launch. This article is my developer report after four years: what I keep appreciating, what still annoys me, and what I recommend to client teams that are still hesitant to deploy it in 2026.

What Copilot Is, Concretely

GitHub Copilot is a code completion and generation service based on language models. It proposes real-time suggestions as I type, and also responds to natural language questions through a chat interface. Published by GitHub and Microsoft since 2021, it has become the de facto standard among the majority of developers I meet on missions.

The tool integrates into the main editors: Visual Studio Code (the most polished integration, the one I use), Visual Studio, JetBrains (IntelliJ, PyCharm, WebStorm), Neovim, Xcode, Eclipse. I mainly work in VSCode and IntelliJ depending on projects, and the experience is consistent between the two.

The Features I Actually Use

Several surfaces I've integrated into my flow. Inline code completion while typing, accepted by Tab — that's what saves the most time on boilerplate. Copilot Chat for explaining legacy code, suggesting refactors, or generating tests. Slash commands (/explain, /fix, /tests, /doc) for targeted contextual actions. Copilot in the CLI for crafting exotic terminal commands. Pull request summaries automatically generated on GitHub. And more recently Copilot Workspace for planning and implementing multi-file changes.

Multi-Model Opening, the Big 2024 Change

Since 2024, GitHub Copilot is no longer tied exclusively to OpenAI models. I can now choose the engine by task: OpenAI's GPT models (historical), Anthropic's Claude models, Google's Gemini models. This was a major change I welcomed with relief — I regularly switch to Claude for complex refactors and reasoning, and keep OpenAI for fast completions. This opening lets me optimize the access conditions/quality ratio by prompt type.

Access conditions

GitHub Copilot offers several tiers. open access: limited quotas, available for all GitHub accounts since 2024 — an excellent lever to try it out. Pro: individual access plan, higher quotas, access to advanced models. Business: for teams, with organization controls. Enterprise: knowledge base integration, chat on the organization's source code, advanced controls.

Worth knowing: students, teachers, and maintainers of popular open source projects get full open access access via GitHub Education. I systematically send the link to young developers asking how to start.

Code Confidentiality: The First Point I Check

This is one of the points most scrutinized by the clients I work with. Here are the official commitments I verify in the documentation. In Business and Enterprise versions, prompts and suggestions are not used to train models. In the individual version, the user can disable this telemetry in settings. An optional filter can block suggestions matching public code, limiting license risks (useful for codebases that don't want to inadvertently absorb GPL code). Organizations can apply usage policies to their entire team.

For sensitive codebases (healthcare, defense, finance), Business and Enterprise versions are the only options compatible with most internal policies I've seen. The individual version remains acceptable for open source or personal code.

Use Cases Where Copilot Really Shines

Based on my experience and the teams I accompany. Boilerplate: generating getters/setters, validations, Zod or Pydantic schemas — massive time savings. Unit tests: rapid writing from an existing implementation, provided you systematically review the test cases. Cross-language conversion: moving from a Python script to a TypeScript equivalent, for example. Inline documentation: JSDoc, Python docstrings, function comments. Learning: exploring a new framework by asking natural language questions about the code I'm reading.

Limits I Still See in 2026

Copilot remains an assistant, not an autonomous developer. A few points I repeat in every training session. It can generate code that compiles but doesn't follow project conventions — you need a strict linter alongside. On non-standard legacy code, suggestions quickly lose relevance. API hallucinations happen: invented methods, nonexistent parameters, uninstalled libraries. Context is limited: on large multi-file refactors, the human remains essential to steer.

My rule for all my teams: anything Copilot proposes must go through the same code review as a human commit. No shortcut, no exception.

Credible Alternatives I Test

The AI developer market has expanded. My main alternatives. Cursor: VSCode-based editor with more deeply integrated AI, I use it for greenfield projects. Cody (Sourcegraph): focus on indexing large codebases, excellent for monorepos. Tabnine: self-hostable models, on-premise option for sovereignty constraints. Codeium: generous open access version, multi-IDE. JetBrains AI Assistant: native integration into the JetBrains ecosystem, worth considering if you're 100% JetBrains.

My Trust-Vault Reading

On my Trust Score grid, GitHub Copilot benefits from several solid assets. Security: backed by Microsoft, SOC 2 Type II, enterprise compliance. Transparency: public access conditions, complete API documentation, regular changelog. Reliability: completion quality recognized by a majority of developers surveyed in the Stack Overflow Developer Survey 2024. Privacy: Business and Enterprise versions compatible with GDPR requirements.

The main attention point remains the dependency on a single cloud provider (Microsoft) and the need for a clear usage policy for teams handling sensitive code. For most teams I accompany, Copilot Business with public code filter enabled remains the most pragmatic combination.

--- Sources: GitHub Copilot official documentation; Stack Overflow Developer Survey 2024; AICPA SOC 2 framework; Anthropic Claude on Copilot announcement 2024; Microsoft Trust Center.