Hacker News Reader

Hacker News

Comment by ttrmw | Hacker News Reader

Parent Full thread

ttrmw

what're you finding better than cursor now?

View on HN

Claude Code with Pro, Max100, or Max200 subscriptions. Works with any IDE including none.

For the time being, nothing comes close, at least for me.

Can you please share your Claude usage workflow?

I use Github copilot and often tend to be frustrated. It messes up old things while making new. I use Claude 4 model in GH CP.

I use github copilot chat right now. First I use ask mode to ask it a question about the state of the codebase outlining my current understanding of the condition of the code. "I'm trying to x, I think the code currently does y." I include a few source files that I am talking about. I correct any misconceptions about the plan the llm may have and suggest stylistic changes to the code. Then once the plan seems correct, I switch to agent mode and ask it to implement the change on the codebase.

Then I'll look through the changes and decide if it is correct. Sometimes can just run the code to decide if it is correct. Any compilation errors are pasted right back in to the chat in agent mode.

Once the feature is done, commit the changes. Repeat for features.

Does it remember context from chat mode and when you switch to agent mode?

Yes. I think it used to be separate tabs, but now chat/agent mode is just a toggle. After discussing a concept, you can just switch to agent mode and tell it to "implement the discussed plan."

Yes, it can't change between edit and ask/agent without losing context but ask <-> agent is no problem. You can also change to your custom chat modes https://code.visualstudio.com/docs/copilot/chat/chat-modes without losing context. At least that's what I just did in VSCode Insiders.

Here are some nice copilot resources: https://github.com/github/awesome-copilot

Also, I am using tons of markdown documents for planning, results, research.... This makes it easy to get new agent sessions or yourself up to context.

I also do the same. I am on the 200$ maxpro plan. I often let the plan go to pretty fine level of detail, e.g. describe exactly what test conditions to check, what exact code conditions to follow. Do you write this to a separate plan file? I find myself doing this a lot since after compaction Claude starts to have code drift.

Do you also get it to add to it's to-do list?

I also find that having the o3 model review the plan helps catch gaps. Do you do the same?

GitHub Copilot models are intentionally restricted, which unfortunately makes them less capable.

I'm not the original poster, but regarding workflow, I've found it works better to let the LLM create one instead of imposing my own. My current approach is to have 10 instances generate 10 different plans, then I average them out.

Github Copilot is weirdly bad, and all the alternatives are better. Sometimes people think "they have the same model, must be the same," but it's not.

My problem with Claude code versus Cursor is that with Cursor I could "shop around" the same context with different foundational model providers, often finding bugs this way or or getting insights.

Sometimes one model would get stuck in their thinking and submitting the same question to a different model would resolve the problem

I’m unaffiliated, but I’ve really been enjoying this - https://github.com/BeehiveInnovations/zen-mcp-server

It allows you to have CC shoot out requests to o3, 2.5 pro and more. I was previously bouncing around between different windows to achieve the same thing. With this I can pretty much live in CC with just an editor open to inspect / manually edit files.

This was my answer as well. And I think it just highlights all the serious dangers for the "API wrapper companies" compared to the foundation model companies.

User experience is definitely worth something, and I think Cursor had the first great code integration, but then there is very little stopping the foundation model companies from coming in and deciding they want to cut out the middleman if so desired.

Amp Code is also very good, they released about 2 weeks ago their Oracle feature which leverages o3 to do reviews (https://ampcode.com/news/oracle). Amp leans the closer to Claude Code more than other solutions I’ve seen so far, the team there is really leaning into the agentic approach.

I watch the changes on Kilo Code as well (https://github.com/Kilo-Org/kilocode). Their goal is to merge the best from Cline & Roo Code then sprinkle their own improvements on top.

Cline is absolutely fantastic when you combine it with Sonnet 4. Always use plan mode first and always have it write tests first (have it do TDD). It changed me from a skeptic to a believer and now I use it full time.

How much is it costing you?

As much as you theoretically want to spend, since it's pay-per-use.

I spend $200/month by using Sonnet 4. Could be higher if you want to use Opus.

You can use Claude Code as a provider if you want it subscription based

https://docs.cline.bot/provider-config/claude-code

I use Roo Code (Cline fork) and spend roughly $15-30/mo by subscribing to Github Copilot Pro for $10/mo for unlimited use of GPT-4.1 via the VS Code LM API, and a handful of premium credits a month (I use Gemini 2.5 Pro for the most part).

Once I max out the premium credits I pay-as-you-go for Gemini 2.5 Pro via OpenRouter, but always try to one shot with GPT 4.1 first for regular tasks, or if I am certain it's asking too much, use 2.5 Pro to create a Plan.md and then switch to 4.1 to implement it which works 90% of the time for me (web dev, nothing too demanding).

With the different configurable modes Roo Code adds to Cline I've set up the model defaults so it's zero effort switching between them, and have been playing around with custom rules so Roo could best guess whether it should one shot with 4.1 or create a plan with 2.5 Pro first but haven't nailed it down yet.

Looking at Cline, wondering what the real selling points for Roo Code are. Any chance you can say what exactly made you go with Roo Code instead of Cline?

Cline has two modes (Plan and Act) which work pretty well but Roo Code has 5 modes by default. (Code, Ask, Architect, Orchestrator, Debug) and it's designed so that users can add custom modes. e.g. I added a Code (simple) mode with instructions about the scale/complexity of tasks it can handle or decide to pass it to Code for a better model. I also changed the Architect mode to evaluate whether to redirect the user to Code or Code (simple) after generating a plan.

Roo Code just has a lot more config exposed to the user which I really appreciate. When I was using Cline I would run into minor irritating quirks that I wished I can change but couldn't vs. Roo where the odds are pretty good there are some knobs you could turn to modify that part of your workflow.

Devin is light years ahead of Cursor. It’s not even the same category!

I stopped writing code by hand almost entirely and my output (measured in landed PRs) has been 10x

And when I write code myself then it’s gnarly stuff and I want AI to get out of my way…so I just use Webstorm

Reply on news.ycombinator.com