Cursor vs VS Code (2025): Why I’m Back with VS Code’s AI

I’ve been on VS Code for six years. I tried Cursor for about a year to see if an AI-first editor could replace my daily setup. Cursor was bold and fast, but a few things pushed me away: default hotkeys like Cmd+K and Cmd+L collided with my muscle memory, the UI changed a lot, and the “manual/edit” flow never felt steady. To Cursor’s credit, when I burned through my free AI credits in the first few days, they topped me up so I could keep going.

This summer I circled back to VS Code + GitHub Copilot Chat and it’s grown up: a clean Chat panel, a predictable Edit mode with diffs, and a one-click Generate Instructions that onboards AI to my codebase. For solo work, it fits.

---

TL;DR

• I’m back on VS Code for day-to-day work.
• Edit mode in Copilot gives safe diffs I can review.
• Generate Instructions onboards the AI to my project in one step.
• I like that I can wire OpenRouter into VS Code’s model lineup via BYO key/provider.
• Cursor still wins when I want aggressive, repo-wide, agent-style changes.
• Pricing reality: Cursor has Pro $20, a not-listed Pro Plus $60, and Ultra $200. Copilot uses a usage-based model with monthly allowances for premium requests.

---

What I liked (and disliked) last year

Cursor highs: strong whole-repo awareness, aggressive multi-file edits, and real Agent workflows that can run commands and touch many files. You can add project rules so the AI follows your style.

Cursor lows (for me): hotkey collisions, fast-moving UI, and edit flows that never settled into my hands. Personal friction, not everyone’s.

VS Code: stayed familiar, then quietly added the AI features I wanted — without hijacking my habits.

---

2025 snapshot (solo-dev lens)

Models & “brains”

• Cursor lets you pick frontier models (OpenAI / Claude / Gemini, etc.). It references GPT-5 usage in docs, so it’s wired for the latest.
• VS Code (Copilot) has a model picker and supports bring-your-own provider keys. I like that OpenRouter works inside VS Code via the ecosystem (official BYO + marketplace providers such as OpenRouter/Continue/CodeGPT). That flexibility matters to me.
• Net: model quality is no longer the differentiator. It’s about how smoothly the tool fits your flow.

Chat + Edit

• VS Code: Chat panel is calm and predictable. Edit mode = select files/regions → prompt → see a diff → apply. Low-drama.
• Cursor: does wide edits fast and has Agent modes for bigger jobs. Great when you want the AI to roam and make sweeping changes.

“Make the AI understand my project”

• VS Code: one click → Generate Instructions. It scans your workspace and creates a copilot-instructions.md tuned to your structure and conventions. Tweak it once; suggestions follow your house rules.
• Cursor: add rules under .cursor/rules to encode style, patterns, and domain knowledge. Power-user friendly and very granular.

Autonomy knobs

• VS Code: Copilot coding agent can run tasks in the background; delegate from chat or GitHub’s Agents panel. Keeps human-in-the-loop review.
• Cursor: the Agent is built for sweeping, multi-file work and shell commands. High leverage if you’re comfortable giving it initiative.

Pricing (solo reality)

• Cursor: public page shows Hobby (free), Pro $20/mo, Ultra $200/mo. There’s also a Pro Plus tier not listed on the pricing page; community/docs mention Pro Plus and users report it at $60/mo. If you live in daily agent territory, expect spend in that ballpark.
• Copilot (VS Code): usage-based model with monthly allowances that reset. You get a base allowance for code completions and chat, with premium requests tracked separately. Additional premium requests can be purchased if you exceed your allowance.ion: “Six years on VS Code, one year with Cursor. 2025 comparison of AI chat, Edit mode, model choice (incl. OpenRouter), project onboarding, and pricing.”

---

Quick comparison (as of Sept 1, 2025)

Setup speed

• VS Code (Copilot): Enable Copilot → Generate Instructions → chat/edit
• Cursor: Install → pick models → optional .cursor/rules for deep control

Edits

• VS Code (Copilot): Edit mode shows diffs you approve; conservative by design
• Cursor: Agent can change many files and run commands; high leverage

Project context

• VS Code (Copilot): Auto-generated copilot-instructions.md from workspace
• Cursor: .cursor/rules for scoped style/architecture/domain rules

Model control

• VS Code (Copilot): Model picker + BYO keys; OpenRouter fits via provider
• Cursor: Explicit picks; supports newest families

Cost feel

• VS Code (Copilot): Usage-based with monthly allowances; track premium request percentage
• Cursor: Tiered, usage-aware mindset; Pro Plus $60 exists though unlisted

---

How I actually use them

• Daily pairing: VS Code Chat for Q&A, quick refactors in Edit mode, and generated instructions to keep suggestions on-brand for my codebase. Minimal friction.
• Occasional big chores: when I want a brave repo-wide change or a scaffold, Cursor’s Agent still feels like a rocket. I supervise.

---

My verdict (for now)

I’m back on VS Code. Copilot’s Generate Instructions, Edit mode, and calmer UX match how I work. When I want a big autonomous push, I still spin up Cursor’s Agent. Day-to-day, VS Code wins for me.