Ask HN: How much better are AI IDEs vs. copy pasting into chat apps?

I'm a staff software engineer doing a mix of front-end and back-end with emphasis on front-end.

I use both Cursor on Claude 3.7 and ChatGPT on 4o/o3. Cursor seems kind of "dumb" compared to 4o, but it's a good workhorse.

I let Cursor handle the basics - basically acting as a glorified multi-file autocomplete. I think through common problems with 4o, tough problems with o3, I copy all of Svelte's docs into 4o (https://svelte-llm.khromov.se/) to get good Svelte 5-focused feedback, I have 4o code-review what Cursor writes from time to time, I have 4o, sometimes o3, generate "precise" prompts that I'll give to Cursor when me talking off-the-cuff to Cursor doesn't get good results after a few attempts.

I don't consider myself an expert in these areas yet so I might be misusing Cursor, or not making enough use of its rules system, or something. I feel like I get good value for my ChatGPT subscription. I don't feel like I get good value for my Cursor subscription, but I also still feel like keeping it because $20 to type a lot less is still pretty nice. I would be upset if I only had a Cursor subscription and no access to ChatGPT. I am pretty hesitant to pay for AI à la carte. I feel much better about working within the limitations of a known subscription cost.

I'm SO surprised to not see aider-chat everywhere mentionned.

I've been using it for about a year and it's incredible.

My take is that most people did not invest the little time necessary to get acustomed to its workflow.

My advices for aider are:

- familiarize yourself with the chat modes (architect, code, ask)

- familiarize yourself with the edit modes (diff, whole, etc) and know which to use for a given model. Indeed not all models handle all modes equally well.

- make the code one feature at a time, by small chunks if needed, by limiting the contxt to the relevant files.

- practice to learn how to best phrase stuff.

- write you coding preferences into aider convention files. Things like "always use type hints, beartype for type checking, python click for the cli, add many comments".

I'm mainly doing python and with proper comments and type hints it's really easy for models to write code that works with the rest of the repo even with only a few files in its context.

I settled on gptel, which is an LLM package for Emacs. It's kind of a mix of both approaches:

1. You have chats right there in the editor, easy to copy/paste and manage context/history without switching to a browser. You can also quickly add files or portions of files to the context or remove them again.

2. You can choose which model you want to use for what, granted you have an API key.

3. You can quickly highlight some code and ask for a change to it, which along with managed context is powerful.

I tried auto complete again and again but it doesn't work for me. At first I think "yeah, that's what I wanted to write", and then I have to look closer to realise it's not, and that completely breaks my flow and concentration. I can always write some pseudo code and proactively convert it to real code, I like to be in the driver seat.

Context management is really central to my workflow, I manage it like a hawk. Models tend to deteriorate as context content increases, in my experience, so I really try to keep it narrow.

For that reason, and because our clients didn't sign up for their code to be sent to Anthropic et al, I _mostly_ use models like I would use StackOverflow, not to generate non-trivial code I'd actually use.

But having the chats in my editor is really invaluable for me. Powerful text wrangling features make a difference in both speed and motivation.

I use it pretty heavily with pretty much only the high-end models and pay about $15 per month.

I use Claude Code (cli tool) and it's on another level. Not even comparable to code autocomplete à la Copilot or copy/pasting into a LLM chat app. It knows about your whole code base, can use external tools, read documentation, run tests, etc. Even as an experienced developer, it's been a huge productivity boost. The main downside is that it can quickly get expensive.

I don't use IDEs to begin with. I use a text editor, since I am a UNIX man. So, I don't even have a reason to look at the "AI" flavor of an IDE. The closest thing to an AI IDE that I have tried out were tools like AIDER and Jack's "goose" agent. Neither of those specialized tools has been satisfactory. They all performed worse than just the LLM IMO. I am sticking to crafting my own context that I supply to the LLM. Tools like Simon W's `llm` tool help A LOT to be more efficient at using LLMs in a daily setting.

agentic ides are better for vibe coding because every time prompted it has the context of the whole codebase and it can edit multiple files and even run console commands or write docs, its easier to vibe debug too, so its much better than copypasting file by file with something like google ai studio. though agentic ides usually come with an elevated price tag.

My current favorite (and free) workflow:

1. Drop project files into https://files2prompt.com to copy a prompt with all file contents

2. Paste into https://aistudio.google.com/ and set a low (or 0) temperature and low top_p

Since Gemini 2.5 Pro is free in AI Studio at the moment, and there's a 1M token limit, this works for most things I need. Cursor is better in some cases where I need a bunch of small edits across a bunch of different files.

I've been using Windsurf for a few weeks. I'm a novice programmer trying to build a web app using React and NextJS.

The "context-in-the-codebase" thing for AI-based IDEsis overrated IMO. To extract the most from it, you have to remember to @mention the files.

If you don't remember to @mention specific files, the agent simply tries to perform searches (i.e., it has access to that tool) on the files and folders until it gets some results... and will usually keep broadening the search until it does.

It works well enough I suppose. But I still find myself beginning new chats (for example, per feature) because the model still loses its place, and with all the code/lint fixes it does, it starts to lose context.

Then you're right back having to @mention more files to ensure the model knows how to structure your front end, back end, etc.

(Please excuse any misnamed development terms.) :-)

I rolled out Claude Code and Cursor to my whole company. They’re really much better than copy paste. And very affordable compared to professional developer wages.

Wrote about it: https://ghiculescu.substack.com/p/nobody-codes-here-anymore

> The issue with Copilot is that it's not as smart as the "thinking" chat apps.

Upgrade to a paid plan and you can use many of the same “thinking” models.

Honestly, your entire question is best answered by signing up for a free trial of one of these tools and using it. Not the free tier, a trial or a paid plan.

Copy and pasting into another app is extremely inefficient. You really need to try a tool that integrates your model of choice into the editor.

I've finally settled on this:

- A custom vscode plugin to help me copy file contents from the tree view, along with file paths

- A chat based ide (LibreChat, but anything will do)

- An agent that syncs code back, once I'm happy with the conversation (https://github.com/codespin-ai/codebox)

Sometimes I add the agent right at the beginning. Sometimes I do that when I feel the code is ready to be written back. Another thing with the codebox agent is that it lets the agent run unit tests (even inside a docker compose environment, if you want integration tests); so in some cases it can be fire-and-forget.

I use copilot through VSCode. The copilot plugin lets me either ask questions or ask for edits to my code. I can choose which files to include as an additional context to the prompt. I can additionally choose which model to use. Some models are unlimited, others have some monthly quota. GPT-4o is unlimited and quite good with short tasks.

It has also shortcuts for instance fixing errors in the code by just selecting "fix with copilot". Its a bit hit or miss but for simple things work well.

It is very good and seamless. Cannot think of ever copy pasting multiple files of code to some web chats and back.