VSCode Is Losing Developers to Simpler Tools — and AI Might Be Why
A developer posted on HN: “I feel like VSCode is falling apart — and not just VSCode, but other pieces of software I always use.” The responses revealed something interesting: many developers are quietly moving to lighter alternatives like Zed, Cursor, and, in some cases, plain terminal editors. The common thread isn’t just performance — it’s that VSCode’s pivot toward AI features has made the core editing experience worse.
The Complaint Nobody Expected to Care About
The original post was modest — 8 points, 7 comments. But the thread resonated.
The developer’s core observation: software quality seems to be getting worse, not better, even as AI promises to accelerate development. They weren’t talking about one specific bug. They were describing a general sense that the tools they rely on feel less reliable than they used to.
Responses ranged from specific grievances to philosophical:
– “I have an older VSCode. A few versions ago, it stopped connecting to my remote machine.”
– “The future of VSCode is quite uncertain, as newer IDEs are moving toward not seeing any code — and soon not even seeing the filetree.”
– “I simply don’t understand the new VSCode Agents window. Many of us are returning to the simplicity and power of the terminal.”
That last one is the most telling. The “Agents” window is VSCode’s big AI integration — and it’s alienating the developers it was supposed to help.
What Actually Changed
VSCode’s transformation has been gradual but clear:
2019-2022: VSCode was the lightweight, fast, extensible editor that replaced Atom and Sublime Text for most developers. Its strength was that it stayed out of your way.
2023-2025: AI features started arriving. GitHub Copilot, then Copilot Chat, then the Agents panel. Each addition added weight. Extensions multiplied. Startup times grew. The “simple editor” identity started fraying.
2026: The button in the upper right corner — the one that transforms VSCode from a code editor to what one commenter called “a Claude/Codex clone” — became the visual symbol of the shift. Clicking it feels like leaving your editor and entering a different product.
The AI Integration Problem
Here’s the core tension: AI coding tools need to understand your codebase to be useful. That means:
– Indexing your files
– Running background processes
– Consuming memory
– Sending code to external APIs
None of that is free. Every AI feature added to VSCode comes with a tax on the editing experience.
For developers who don’t use AI features, that tax is pure downside. They get the weight without the benefit.
For developers who do use AI features, the experience is mixed. The integration is powerful, but the workflow feels less like “writing code” and more like “managing an AI that’s writing code.”
What Developers Are Switching To
Zed came up repeatedly. Built by the original authors of Atom, Zed is explicitly designed as a lightweight alternative — fast startup, minimal UI, native AI integration that doesn’t bloat the base experience. One commenter said, “It always felt heavy on my older machine. Zed works okay.”
Cursor — positioned as “VSCode but with AI-first design” — attracted developers who want AI as the primary workflow, not an add-on.
Back to terminal — several developers described returning to vim, emacs, or just tmux + a good syntax-highlighted editor. The appeal is control: no background processes, no indexing, no mysterious memory usage.
The Real Story Behind “Falling Apart”
Software really is getting harder to rely on in some ways. Not because developers are getting worse — but because:
AI features introduce non-determinism. A traditional bug is reproducible. An AI that gives bad advice or suggests wrong code is harder to debug because the suggestion changes every time.
Integration complexity compounds. Modern software is layers of APIs, cloud services, and AI models. When something breaks, it’s not always clear which layer is responsible.
Update: velocity has increased. AI features are moving faster than traditional software development processes can validate. Features ship, get refined, and sometimes get deprecated quickly. That creates a sense of perpetual beta.
None of this means AI tools are bad. It means the tooling ecosystem is in a transition period — and the tools that handle that transition gracefully will win developer loyalty.
