What's great
I have been using ChatGPT as my primary LLM for the last 2 years. I had heard of Claude during that time, of course, but never really explore it. Finally, I became tired of ChatGPT's tech-bro style of writing and its sycophancy. So I tried Claude and woah, what a difference! It's quite something to interact with an LLM that sounds a feels more human, and yet retains (of perhaps exceeds) the technical capabilities. Also Cowork and Claude Code are at another level. Recommending to everyone I know who use other LLMs to switch to Claude now.
What needs improvement
Token limits
vs Alternatives
Claude is simply more capable, more philosophical and deep thinking, and overall feels more like speaking to a very capable, intelligent, yet also considerate, employee.
How accurate is web search and are citations reliable?
In my experience, very reliable
Does the 200k+ token context stay coherent in practice?
Completely. Much more so than other LLMs I've used
What's great
Tried lovable and loved it. Incredibly easy to use, and the outputs are genuinely good. I like that you can toggle to pause deploy and instead focus on providing more context so that the output is more aligned with your vision.
vs Alternatives
Was looking at other website vibecoding options but loveable seemed to have the best value for money, and the best user interface.
How often do you encounter agent loops or timeouts?
Did not encounter any when using to design my website.
Does it work best for greenfield rather than legacy projects?
Maybe, not sure, as I started a new project from scratch.