Ghost text, and what it is actually reading
The original capability is inline completion — greyed-out code ahead of the cursor, accepted with Tab.
What determines suggestion quality is context, and Copilot’s context is narrower than people assume. It reads the current file, related open tabs, and nearby symbols. It does not index your repository. A function defined in a file you have not opened is, for practical purposes, invisible to it.
That explains the most common complaint — suggestions that reinvent a helper you already have — and it explains a practical habit worth adopting: open the files whose conventions you want followed before you start writing. Users who do this report markedly better suggestions, and it is the single highest-return adjustment available.
It also explains where Cursor and Cody differ: both index, and both are stronger on questions that span a codebase.
Chat, agent mode, and where the boundary sits
Copilot Chat is a panel that can see your selection, file or workspace — for explaining unfamiliar code, generating tests, or proposing a fix for an error. Slash commands cover the common requests, and @workspace broadens the context at some cost in precision.
Agent mode plans and executes multi-file changes, running commands and reading their output. It is the same pattern as Cline or Claude Code, and it arrived later than in those tools — worth knowing when comparing against reviews written before it shipped.
Availability of both varies by plan and by editor, and that variation is a real source of confusion: two developers on different tiers have genuinely different tools while calling them the same name.
The GitHub integration, which is the part competitors cannot copy
Copilot reaches beyond the editor into the platform. Pull request summaries are generated from the diff. A coding agent can be assigned an issue and open a pull request against it. Code review suggestions appear in the PR itself.
For a team already on GitHub, that closes a loop no editor-only tool can reach. It is also the strongest argument for choosing Copilot over a better editor experience: the value is not only in the typing.
Data handling, and what the tiers actually differ on
Code is sent to hosted models. That is unavoidable and true of every tool in this category except the self-hosted ones.
What varies is retention and training. GitHub documents that Business and Enterprise plans do not retain prompts or suggestions for training, while individual plans have historically offered that as a setting. There is also a filter that blocks suggestions matching public code, which mitigates — without settling — the question of code that resembles its training data.
Read GitHub’s current documentation rather than any summary, this one included. These terms are exactly what changes between plan revisions, and an organisation adopting Copilot should be reading the source before it signs.
What no tier offers is local execution. If code cannot leave your network, this is the wrong category — see Tabnine for the air-gapped option and accept the capability trade.
What it costs to run, in setup and in money
- A supported editor — VS Code, Visual Studio, JetBrains IDEs, Neovim, Xcode and others. The broadest coverage in this category by a clear margin.
- A GitHub account and a Copilot plan. A free tier exists with monthly limits, plus paid individual, business and enterprise plans.
- Verified students, teachers and maintainers of popular open-source projects can qualify for free access. Check eligibility before paying — a substantial number of people who pay do not need to.
- An internet connection; suggestions are generated remotely and latency is noticeable on a poor link.
- For organisations, an administrative decision — Copilot is configured at the org level, including policy on public-code matching and which members have access.
Getting real value out of it
Three habits separate developers who find it transformative from those who find it mildly useful.
Open the relevant files first, as above. Context is what it reads, and you control that.
Write the comment before the code. A precise comment describing intent produces markedly better completions than an empty line, because you have given the model the specification.
Review everything in unfamiliar territory. Suggestions are most confident where they are least reliable — an unfamiliar language or API is exactly where plausible-wrong code is hardest to catch and most expensive to ship.
Who it is right for
Almost any professional developer, which is the point rather than a hedge. It is the lowest-friction entry to AI-assisted coding: install an extension and carry on.
Teams already on GitHub, where the PR and issue integration compounds the editor value.
Polyglot developers and anyone using an editor the newer tools ignore — JetBrains, Neovim and Xcode users have far fewer options.
Enterprises needing org-level policy, audit and documented data handling, where a startup’s terms would not survive procurement.
It is a weaker fit for developers who want to choose their own model, for anyone needing local execution, and for those whose work spans a large codebase where indexing tools have a real edge.
Why teams standardise on it
- No workflow change — your editor, keybindings and extensions stay exactly as they are.
- The broadest editor support in this category.
- Platform integration — pull requests and issues, not only files.
- Enterprise controls: org policy, audit, documented handling, and a procurement path that already exists.
- Predictable per-seat billing rather than variable token spend.
- Free access for students, teachers and OSS maintainers.
Where it falls short
- No model choice, no local option. You use what the plan provides.
- No repository index, so it misses code you have not opened — the most consequential functional limit.
- Confidently wrong suggestions that are hardest to catch in unfamiliar languages.
- Feature availability varies by plan and editor, which makes comparing experiences difficult.
- Licensing questions persist around training on public repositories; the duplication filter mitigates rather than resolves.
- Latency on a poor connection makes completion actively annoying.
Leaving is cheap
Uninstall the extension. Your code, repository, editor and settings are unaffected — nothing about your project depends on Copilot having been there.
That is worth stating because it is the opposite of the editor-fork tools, and it lowers the cost of trying it. There is no migration, no export, and nothing stranded.
Alternatives worth comparing
- Codeium — comparable completion with a genuinely free individual tier.
- Cursor IDE — deeper integration and codebase indexing, at the cost of switching editors.
- Tabnine — when self-hosting and licence provenance are the requirement.
- Cline — agentic work in VS Code with your own API key and model choice.
Compiled from GitHub’s Copilot documentation and public sources. Plan features and data-handling terms change — verify against GitHub’s own documentation. We have not hands-on tested this tool. Last reviewed 16 August 2026.
