What do you NOT like about Cursor / VSCode / Claude Code desktop / Codex / etc.?

1•nokturn•1h ago

I am building a highly integrated, cross-provider agentic workstation (its neither an IDE nor an ADE - does a bit of both, with additional unique features on top), and I would love for you guys to rant about what you hate about the tools you currently use. You don't even have to mention the tool name if you want.

Examples: - "X is cool but lacks worktrees support" - "I hate that I have to go to GitHub to check if my PR has been already approved" - "I wish that X would support remote sandboxes"

Things that either are an annoyance to you, time wasted, or things that you think would make you more productive.

Background of what my tool aims to do: As a software developer, make sure that the only moments your attention is required is when there are __important__ questions to answer or when its time to validate the work. No bouncing between jira/linear <-> IDE <-> agent terminals <-> github.com etc.

Any pain points you guys can share will be extremely valuable, and I'll be grateful for all the time spent.

Comments

pitched•48m ago

They all do way too much. I just want it to load fast, type with no latency, and get the hell out of the way. Like vim, but less confusing.

jdabney•41m ago

I also want them to not take up GBs of RAM for only having a few windows open. I want something not based on Electron. I want something native instead of written in Javascript.

nokturn•32m ago

Thanks for bringing the Electron topic! I'm actually using Electron, and it's mostly a web app - the reason being that will make it easier to host it online when I add support for cloud sandboxes. I imagine in the near future that we (developers) won't be writing much manual code anymore, and our time will be spent validating work or guiding agents to do said work (planning, architecture decisions, validating business logic, etc).

Electron on its own isn't that bad, and surely not responsible for GBs of RAM - the main bottleneck I'm facing when dealing with a dozen+ parallel tasks is that each claude code terminal eats up easily 200MB of RAM.

When you mention Electron, is it because in most Electron apps the experience doesn't feel snappy, as it would vs native?

mtmail•15m ago

soloterm.com is based on Tauri and uses less memory. It might not be your bottleneck but memory usage adds up with many projects and virtual machine. Not everybody wants to run their virtual machines in the cloud.

d3Xt3r•21m ago

I also don't want any of its dependencies to be written in Javascript. I also want minimal, auditable dependencies - build, package and runtime. Ideally, I want the app to be a single, portable binary that I can run anywhere, without installing. I also don't want the app to pollute my filesystem and litter configs and other dependencies everywhere. I want the app to respect the Freedesktop standards.

I also don't want the app to use GTK, libadwaita and the like. Although Qt is somewhat acceptable, I would strongly prefer something more lightweight like Slint, Iced or egui. Or even better, make it a terminal app with a TUI (that doesn't use bloated frameworks like Ratatui/Bubbletea etc).

nokturn•8m ago

Many thanks for taking the time to share mate. - no javascript (you're the #2 mentioning this) - small footprint, auditable (i read opensource is a must?) - no outside dependencies - Linux as a first-class citizen - TUI > GUI

I can read that you're probably a terminal power user, so how do you keep track of non-coding stuff like tracking JIRA tasks and handling code reviews? Are you using MCPs for that, or a TUI orchestrator with support for those sort of things - or do you manually go to those platforms yourself?

Once again, thank you for taking the time to discuss these things :+1:

nokturn•28m ago

- performance - simple/objective

Noted! Thank you first of all!

Besides those two points, is there anything else you think could save you time? Ex. notify you when your PR got approved, auto-fix CI errors, not ping you every 10 seconds asking permission to execute a bash command again, send a msg notification somewhere when work is done, or being able to continue working on your phone, stuff like that

dlcarrier•31m ago

Mostly the bloat. It's not just the order of magnitude more RAM and CPU usage than any reasonable editor uses, it's the lag that is really grating.

Yes, compiling can take time, that's reasonable. What isn't reasonable is for the user interface to be slower than I am. My computer can perform tens of billions of operations per second. When I click on something, there's no technical reason it couldn't respond in at least a 30/th of a second, if not a 60th. There's plenty of software that's been around for decades that can do this just fine.

nokturn•19m ago

Thanks for sharing! Noted! - performance

On a personal note, i totally understand, and I also feel frustrated when I have to wait for the machine, or when i start to feel the laptop getting warm on my lap. So to expand on this a little, would you say a 'monitor' with current CPU/RAM usage would be helpful to at least establish a relation between interactions / resource usage? For example, if you have a dozen worktrees nested inside your main repo and you open cursor, it will immediately jump to 400% CPU just from indexing all those files. Surfacing where this CPU usage is coming from could be a starting point?

mtmail•13m ago

One step back might be to figure out of the target use has dozens worktrees. I never used worktrees for example

nokturn•3m ago

worktrees are a must if you want to run parallel tasks in the same codebase... Either that or sandboxes :)

That example i gave above with the worktrees happened to me recently after a long time of not opening cursor - huge CPU usage and mbp fans immediately firing up. Didn't take long to understand what was happening, but still, no easy way for users to know if its indexing, extensions, or something else entirely.

AEndrix03•8m ago

Clients keep getting heavier and heavier, loading an overwhelming number of tools, while I probably don’t use a good 80% of them. I’ll admit they are gradually improving, but it still feels like things are being unnecessarily overcomplicated, when in reality all you really need is a simple space to write and a send button.

Ancient Secrets

The April every AI plan broke

CRUD Is Broken

Today-dsa – a local-first engine that tells me what to study today

Jon Caramanica is a bad cliché

Why Dunkin' Donuts Failed in India

RAG Eval Comparing Vertex/Bedrock/Azure/OpenAI

Codex Pets for People in a Hurry

Graft – semantic memory for AI agents, without the LLM

Onboarding with LLMs

A Wild Ghost Blog Appears

Thinking Machines: AI that can respond during the interaction

Can we code our way out of gentrification?

Mistral AI's NPM package was compromised

Senior Product Engineer

Brain scans reveal how people with autistic traits connect differently

Linux bitten by second vulnerability in as many weeks

Show HN: ProvisioningIQ – Continuous Benchmarks of AWS/Azure/GCP Latency

Postmortem: TanStack NPM supply-chain compromise

Anxiety vs Depression

Legacy preference bans may not increase college diversity, researchers say

Designing GPUs for Developers: A Conversation with Godot

GM just laid off IT workers to hire those with stronger AI skills

Intel Reenters DRAM Race? A Closer Look at the Z-Angle Memory

Hermes Agent Background Computer Use on macOS via SkyLight Private SPIs

When My Father's Canary Flew Away

Star Wars: Fall of the Empire by Stern Pinball

Why LLMs Make Learning to Code More Important, Not Less

What do you NOT like about Cursor / VSCode / Claude Code desktop / Codex / etc.?

Show HN: NodeDB – High Perfomance Multi-Model Database

Ancient Secrets

The April every AI plan broke

CRUD Is Broken

Today-dsa – a local-first engine that tells me what to study today

Jon Caramanica is a bad cliché

Why Dunkin' Donuts Failed in India

RAG Eval Comparing Vertex/Bedrock/Azure/OpenAI

Codex Pets for People in a Hurry

Graft – semantic memory for AI agents, without the LLM

Onboarding with LLMs

A Wild Ghost Blog Appears

Thinking Machines: AI that can respond during the interaction

Can we code our way out of gentrification?

Mistral AI's NPM package was compromised

Senior Product Engineer

Brain scans reveal how people with autistic traits connect differently

Linux bitten by second vulnerability in as many weeks

Show HN: ProvisioningIQ – Continuous Benchmarks of AWS/Azure/GCP Latency

Postmortem: TanStack NPM supply-chain compromise

Anxiety vs Depression

Legacy preference bans may not increase college diversity, researchers say

Designing GPUs for Developers: A Conversation with Godot

GM just laid off IT workers to hire those with stronger AI skills

Intel Reenters DRAM Race? A Closer Look at the Z-Angle Memory

Hermes Agent Background Computer Use on macOS via SkyLight Private SPIs

When My Father's Canary Flew Away

Star Wars: Fall of the Empire by Stern Pinball

Why LLMs Make Learning to Code More Important, Not Less

What do you NOT like about Cursor / VSCode / Claude Code desktop / Codex / etc.?

Show HN: NodeDB – High Perfomance Multi-Model Database

What do you NOT like about Cursor / VSCode / Claude Code desktop / Codex / etc.?

Comments