andy shi

@adxtyahq This is counter intuitive. In sofeware development, it's been tested agile is better than waterfall. Now to save tokens, you are advocating the opposite. But is the final outcome, the product or sevice developed by AI agent really better using waterfall approach?

@windsurf The new agent is dumb and slow as fxxx. Can you revert to the old cascade ?

Wave 12 is here, and it’s a big one! 📚 DeepWiki-powered docs for every symbol in your codebase 🔍 Vibe and Replace 🐛 100+ bugs squashed 🎨 Brand new UI … and more! Everything that’s new 🧵

@DionysianAgent How much does this cost?

here i'm running 12 - 16 claude 4 opus/sonnet code agents through claude code bridged to eigencode for coordination, additional memory layers & additional tools they're running 95% autonomously with minimal input for several hours straight

@atlantis__labs They changed the UI. This is "Recommended" list. You can choose others.

@iannuttall Yes. Same

me today… anybody else noticing the same?

@zeeg Used cursor for several months. It was soooo slow. Switched to windsurf ever since. These IDEs are not magical. Their capability really depend on the LLM model of choice. But paying enough to get a fast response is important.

If you're not using Cursor - and you're using another editor - you really should reconsider. I keep hearing folks say X is better, or Y is just as good. They're not. You will see this when you actually use them daily. Its evident to me they have the best eng.

@ns123abc Make llm learn from user feedbacks in real time. Not using context.

show me a problem that you think LLMs can’t solve and I'll correctly prompt show you why you're wrong

@aaditsh These are general knowledge. They don't get coverted to tax forms, let alone tax calculations. Something else is needed to make it happen. Who can define those something else?

This guy literally shows how to make AI your accountant in less than 5 minutes

5 year old Asian boy harassed by racist middle school thugs who call him “Ching Chong” and cowardly bully him in Virginia. Please share!

@marclou Same here.

I stopped using sonnet. It felt like coding with a drunk friend—fun until he installs new packages or edits other components “for fun.” My prompt used to be 50% instructions and 50% dont-dos. Switched to gemini-2.5. Finally, some peace.

Traffic in the Sunset has gotten worse: new data shows that congestion on nearby roads has more than doubled since the Great Highway closed to cars on weekdays. 📝: @leahygarrett sfstandard.com/2025/04/07/gre…

@cursor_ai Just used it today. Better than sonnet 3.7 for sure.

Gemini 2.5 Pro is available to all Cursor users! You can enable the full 1M context window if you'd like. We're curious to hear how you think it compares to Sonnet.

@phill__1 @windsurf_ai , ☝️when will this model be available in cascade?

If you haven't tried Gemini 2.5 Pro for coding, you are missing out. Don't just take it from me

@AngryTomtweets Simply not true. So far any merge request I raised using AI without fully understading or checking ended up being rejected. AI makes more mistakes than an intern.

RIP App developers. AI can now build you a professional mobile apps in minutes. Here's how it works: (no coding skills required)

@bsideup @windsurf_ai @cursor_ai You are right. But I wouldn't treat it as Saas. I treat it as an intern. The biggest gain by using it is saving time, not making fancy features. If something I don't know how to do, I would rather chat with perplexity for instructions instead of handing over to these agents.

@andyshi @windsurf_ai @windsurf_ai / @cursor_ai are basically PoCs of AI-enabled development at this moment. - Too much complexity - Low margin - No sensitive default choice of the model (even Claude 3.5 vs 3.7 is a BIG "depends") - UB pricing doesn't work: you consume it multiple times to succeed

@windsurf_ai is just a credit burning machine. You can spend hours fighting the hallucinations without achieving anything but spending $$$. Would be "okay, fine" if not the usage-based pricing. GenAI begs for high margin economics, otherwise it won't survive the wake-up call.

@thekitze Totally true!😂😂

me: hey do this thingie 3.5: no prob sir, done 3.7: i did the thingie. let me also do another thingie. i'm gonna finish all the thingies. omg there are so many thingies to be done in this project. i'm gonna start doing extra thingies. would you also maybe like a drink? let's run npm install drink. fk it let's get crazy up in this b

@cline This is a very bad example of tech advocacy. MCP is not for cline only. If just focus on MCP, you would see better results.

"I still don't understand MCP" is something we hear often from developers. Developers are missing out on the most powerful feature of Cline because the name sounds too complicated. Forget Model Context Protocol - think of it as giving superpowers to your AI assistant. 🧵

@deedydas Perplxity is already doing that. But then it didn't attach enough context when switching.

If we know diff LLMs are good at diff things, why not use the best LLM for the prompt we have? New research from Berkeley and the Databricks founder trained a ~7B model on 1.5M Chatbot Arena ratings to route prompts to the right LLM. It achieves ~1400 ELO, #1 in LM Arena.

@mattyp Can we say MCP is an adapter to other existing api endpoints?

I still have no idea what MCP really is