from Hacker News

Hello everyone, I’m testing and comparing various LLM-assisted coding tools, and I want to know which tool you are currently using in your daily development workflow. Here are some observations and questions I have:

1. Cursor and Windsurf

   - Both work nicely on local, but they use token-saving strategies:

     - With very long context, they may truncate important information, causing the suggested code to miss key details.

     - Even in normal scenarios, complex cases might exceed context or quota limits, interrupting suggestions.

2. “Roo Code” and API-based approaches

   - Directly calling paid APIs (e.g., OpenAI’s ChatGPT/GPT-4 API) works well but is expensive.

   - Some free or community APIs (open-source mirrors, community editions) can be unstable, rate-limited, or slow.

3. Augment Code - It’s said to be one of the most “intelligent” commercial products, but it’s also costly.

   - Many recommend its ability to rewrite, refactor, generate tests, etc., but for simple code completion, its cost-performance ratio may be lower than some smaller vendors or open-source plugins.

4. Refact.ai

   - Listed at the top on SWE Bench, it claims to support code refactoring, generating comments via LLMs, batch rewrites, and more.

   - However, it seems rarely discussed in developer circles. How well does it support?

Questions for the community:

- Which LLM-assisted coding tool are you currently using? (IDE plugin, standalone client, or API-based)

- What are the main reasons for choosing it? (e.g., cost, response speed, context length support, feature set, etc.)

- What pros and cons have you encountered during actual development? Specifically, how does it perform for debugging, refactoring, generating unit tests, automatic bug fixes, etc.?

- If you have switched tools before, why did you switch?

Thank you for sharing your experiences!