March 5, 2026/943 words/5 min read

Stack Check, March '26

How my dev tools finally stopped fighting each other.

Oof, here I go again. I've written about this twice now. I first covered it in The Joy of Keeping Your Hands on the Tools, where I finally figured out that AI-assisted coding didn't threaten the craft I cared about. Later, in Staying Close, Moving Faster, I found a rhythm that lets me draft fast and decide slow. This is the update nobody asked for, but I'm writing it anyway because my stack changed and I feel like blabbing about why.

Claude Code on the Max plan is now my primary tool. For editor work, I'm back in VS Code with GitHub Copilot, and I stopped using Cursor entirely. The reasons involve plenty of frustration. I still believe the best workflow is the one where everything stays out of your way and shuts the hell up when you don't need it.

Cursor's tab completion was, and probably still is, the best in class. After they acquired Supermaven in late 2024, the completions got even sharper and faster, and I built strong muscle memory around them. I could tab, tab, tab through long edits, and the suggestions felt like they were reading my mind! That part was absolutely great, and I miss it dearly.

Everything else decayed. The UI started feeling cluttered and unreliable. Extensions broke in ways their VS Code versions didn't because Cursor is a fork that drifts further from upstream with every update. The cracks showed up everywhere, often in the strangest places. The extension registry felt more and more like a collection of CVEs, and half the things I installed introduced subtle weirdness I'd only notice three days later when something stopped working. I have limited evening coding time, and debugging my editor is a terrible use of it. The editor should disappear and let me do the work I love.

And then they sunsetted Supermaven entirely in November 2025. I understand the business logic and hold nothing against them. They absorbed the team and folded the technology into Cursor's own models, but it still stung! Supermaven was one of the few pure code completion products that did one thing incredibly well. Its flexibility mattered to me, and watching it get absorbed into a single editor's ecosystem felt like losing a tool that respected how I work.

Pure code completion as a standalone product category seems to be dying. Every company wants to turn its product into an agent that owns the whole experience, while the simple "just finish my line intelligently" use case gets bundled into bigger, more opinionated packages. I liked having a sharp knife, especially when everyone wants to sell me the whole kitchen.

So I went back to VS Code with the Copilot plan. It's a tradeoff. The completion quality isn't as good as what Cursor had with Supermaven under the hood. Copilot's suggestions are fine, competent even, but they still land maybe 60% of the time where Supermaven landed 85%. VS Code itself is stable, and its extension ecosystem is battle-tested. I'm happy to give up some completion quality for that. When I'm in the editor, my attention should be on the code instead of a sidebar that rearranged itself.

Claude Code is where the real shift happened, though. In Part 2 it was one of several tools I used for async work. Since then it has become the center of gravity! I hit the Pro limits within my first week, but the Max plan makes the economics work for how I use it. Once I moved to Max, rate limits stopped being something I thought about during a normal evening session. It's worth the money if you use it the way I do.

Claude Code fits how I think about problems. It lives in the terminal and can work across several files without me, so I can hand it a task and go make coffee. When I come back, it has usually tested the change too. The async pattern from Part 2, "go do this and show me what you did", is now my default whenever the work doesn't need me at the keyboard. I still have to set intent and constraints ruthlessly. It mostly does the mechanical work, and I review the diff.

What I didn't expect was how well it works for close, hands-on work too. When I'm stuck on a design question or trying to figure out the right data flow for something new, I can think out loud with it in the terminal. I can also paste an image and show it the problem. That gets closer to having a colleague look over my shoulder than anything else I've used. We argue over the direction for a few rounds, and I usually come away with something I can work with in the editor.

I stay in VS Code with Copilot when I need precise control. When a task can run without me, or when I'm still figuring out the approach, I move to Claude Code. That separation took me three posts and over a year to arrive at, and it's embarrassingly simple in hindsight. I spent too long trying to make one tool cover the whole process. That's how I ended up with a bloated editor that fought me, or an agent that didn't know when to stop.

I'm also more at peace with the fact that this will change again. Whatever I'm using six months from now will probably be different, and that's fine. I want to preserve the handoff between Claude Code's mechanical work and the decisions I make in VS Code. No matter which tools fill those roles, I still review every diff before I ship it.