Cursor and Windsurf are AI-native IDEs (forks). GitHub Copilot is the other procurement gravity well: a plugin family inside VS Code, Visual Studio, JetBrains, Neovim, and increasingly GitHub.com itself (PR, Issues, coding agent). Vol. 21 coding assistants used Copilot as the archetype of ghost text. Vol. 22.1 Microsoft (Phi, Copilot) covers the broader Copilot brand; this lecture is the developer IDE / GitHub SKU.
Vol. 15 HITL still applies when Copilot grows from autocomplete into Chat, multi-file Edits, and async Agents. Vol. 18 deployment via GitHub Actions is often why enterprises standardize here. Next: terminal-native agents—Claude Code.
Learning Objectives
By the end of this lesson, students should be able to:
- Describe Copilot as a multi-IDE plugin + GitHub.com surface, not a VS Code fork.
- Separate ghost-text Copilot from Chat / Edits / Agent (agentic) modes.
- Use content exclusion, org no-train claims, and
.github/copilot-instructionsas privacy/repo controls. - Compare Copilot qualitatively to Cursor, Windsurf, and Claude Code.
- Know when GitHub-centric enterprises should pick Copilot even if a fork wins a demo.
- Keep Vol. 20 copyright/security policy: no exploit tasks, license hygiene on suggestions.
GitHub Copilot is Microsoft/GitHub’s coding assistant: inline completion, chat, multi-file edit, CLI, and (on some plans) repository-aware or async agents that open PRs. It attaches to IDEs you already run and to github.com. Business/Enterprise tiers typically add org policy, content exclusion (block files/repos from being used as context), SSO, and contractual no-train language—verify the current enterprise FAQ. Copilot is HITL: suggestions are not merges.
Plugin vs Fork; Autocomplete vs Agent
| Surface | Where it runs | Pattern (Vol. 21) | Typical buyer reason |
|---|---|---|---|
| Ghost text | VS Code / VS / JetBrains / Neovim / … | Autocomplete | Stay in current IDE |
| Copilot Chat | IDE side panel / inline | Inline chat / Q&A | Explain + edit this file |
| Edits / multi-file | IDE | Multi-file plan | Catch up to fork agents |
| GitHub.com agent / PRs | Browser + Actions | Async agent + HITL review | Issue → draft PR in the forge you already use |
| Copilot CLI | Terminal | Tool-using helper | Shell without switching to Claude/Gemini CLI |
Why a plugin wins
- JetBrains / Visual Studio muscle memory
- One vendor with GitHub Advanced Security / Actions
- Org-wide enablement without new IDE
Why a fork still wins some teams
- AI-native chrome (Cursor/Windsurf)
- Tighter Agent UX in-editor
- Non-GitHub forges (GitLab/self-host git)
Why CLI agents still appear
- SSH / remote / no GUI
- Anthropic or Gemini model preference
- Scriptable agent loops (Vol. 15)
Privacy, Content Exclusion, Repo Context
Copilot’s repo context is a mix of open files, retrieval (where enabled), and optional org indexes. Content exclusion is the enterprise control: paths or repositories that must never be sent as prompt context. Pair with .gitignore-class hygiene and secret scanning. Business/Enterprise no-train language is why many legal teams prefer Copilot over consumer AI IDEs—still read the current contract; do not recite a 2022 blog post.
Vol. 20 copyright: public-code matching filters (when enabled) reduce verbatim regurgitation risk; they do not make suggestions license-clean automatically. Review like any other diff.
Catalog Snapshot (Qualitative)
| Dimension | GitHub Copilot | Cursor | Claude Code |
|---|---|---|---|
| Pricing posture | Seat tiers (individual → Enterprise) plus premium-model / agent request meters on some SKUs—live sheet only | Seats + agent usage | Anthropic plan / API token metering |
| Strengths | Multi-IDE; GitHub PR/Actions gravity; content exclusion; enterprise procurement; Microsoft estate | AI-native Agent UX; VS Code fork | Terminal-native agent; Claude model quality; CLAUDE.md conventions |
| Weaknesses | UX can feel bolted-on vs forks; agentic depth depends on surface/plan; GitHub-centric bias | Not JetBrains-native; not github.com review UI | Not ghost text in JetBrains; token spend on long agent runs |
| Agentic vs autocomplete | Both; autocomplete is the heritage product | Both first-class | Agent-first (little/no IDE Tab) |
| Privacy / repo | Content exclusion + org policy + no-train on paid business SKUs (verify) | Privacy Mode + ignore | Local files + your Anthropic retention settings |
Org Policy Sketch
Educational configuration for repos you maintain. Not an exploit; not instructions to bypass content exclusion.
When to Pick Copilot
Pick Copilot when
- The company is already on GitHub Cloud + Actions.
- Developers span JetBrains, VS, and VS Code.
- Legal wants content exclusion + Microsoft/GitHub DPA.
- Issue → draft PR on github.com is a desired loop.
Pick something else when
- AI-native fork UX wins the bake-off and git is not GitHub-only.
- Anthropic-terminal workflow → Claude Code.
- Google/Vertex alignment → Gemini CLI.
- Non-engineers need prompt-to-app → Bolt / Lovable / Replit.
Related Lectures
| Lecture | Role |
|---|---|
| Coding assistants | Ghost-text product theory |
| Microsoft Phi / Copilot (22.1) | Broader Copilot brand vs this IDE SKU |
| HITL | PR review before merge |
| Deployment | Actions + CI as the ship path |
| Copyright / privacy | Suggestion hygiene + exclusion |
“Copilot is only ghost text.” Chat, Edits, CLI, and github.com agents exist—policy must name modes. Second: content exclusion is automatic without admin setup. Third: Copilot Business no-train means consumer Copilot Chat on github.com is the same contract. Fourth: a fork cannot coexist with Copilot (many teams run both: Tab in Cursor, PRs on GitHub). Fifth: inventing per-seat dollar prices. Sixth: Copilot Agent may merge to main without review.
Knowledge Check
- Short Answer: How does Copilot differ from Cursor at the product-shape level? Answer: Copilot is a multi-IDE plugin + GitHub.com surface; Cursor is an AI-native VS Code fork.
- True/False: Copilot still includes ghost-text autocomplete. Answer: True.
- Multiple Choice: Content exclusion is mainly for: (a) blocking paths/repos from prompt context, (b) speeding Tab latency, (c) replacing unit tests. Answer: (a).
- Short Answer: Name two IDEs Copilot commonly supports besides VS Code. Answer: Any of Visual Studio, JetBrains, Neovim (and expanding surfaces).
- True/False: Copilot Business no-train claims should be verified on the current enterprise FAQ/DPA. Answer: True.
- Multiple Choice: Pricing posture is: (a) seat tiers plus possible premium/agent meters, (b) a memorized $10 figure, (c) GPU-hour only. Answer: (a).
- Short Answer: When is Copilot the default enterprise pick? Answer: GitHub-centric org, multi-IDE developers, and legal wants Microsoft/GitHub exclusion + DPA.
- True/False: A github.com coding agent should merge without human PR review. Answer: False—HITL review + CI.
- Multiple Choice: Vol. 22.1 Microsoft Copilot lecture vs this one: (a) brand/platform vs developer IDE SKU, (b) identical, (c) only Phi SLMs. Answer: (a).
- Short Answer: Which Vol. 21 lecture is Copilot the archetype for? Answer: Coding assistants.
Key Takeaways
- Copilot is plugin + GitHub.com, not a VS Code fork—multi-IDE is the superpower.
- Autocomplete, Chat/Edits, and Agents are different HITL modes; policy must distinguish them.
- Content exclusion and org no-train language drive enterprise adoption; still verify contracts.
- Bake off against Cursor/Windsurf; pick Copilot when GitHub + JetBrains + legal alignment matter more than AI-native chrome.
- Next: Claude Code.
Lab: Enable content exclusion on a dummy path in a GitHub org sandbox (or sketch the YAML). Run one Tab task in JetBrains or VS Code and one Chat edit. If Agent/PR is available on the plan, open a draft PR and require review.
Debate: “We already pay for GitHub Enterprise—must we still bake off Cursor?” Expected: yes, on agent UX; no, if multi-IDE + DPA dominate.
Recap: GitHub Copilot is the enterprise plugin/forge assistant—ghost text plus growing agents, with content exclusion as the privacy lever. Continue to Claude Code.