Get live statistics and analysis of Nityesh's profile on X / Twitter

idk why no one is talking about it but @windsurf_ai gives you almost 4x more ai usage than @cursor_ai for 25% less money ($15 vs $20) Here's the math: Cursor Pro - 500 fast requests (+ unlimited slow requests which in practice don't work) - each AI message, tool call (edit, analyze, search etc) costs 1 fast request with a premium model like Sonnet 3.7 Windsurf Pro - 500 User Prompt credits + 1500 Flow credits - each message you send to AI consumes 1 user prompt credit and each tool call (edit, analyze, search etc) costs 1 flow credit with a model like Sonnet 3.7 so either Windsurf is bleeding through cash here on their ai bills or they've got some neat tricks under the hood that allows them to offer all this while building a sustainable business.

guys i think i may have created the most amazing Claude subagent 😂 @AnthropicAI is going to the moon now 🚀

I've been really impressed with the way Claude Code has started asking questions when you're in Plan mode. So i was curious if i could ask Claude to reveal how this works. I asked Claude - "tell me about the tool you have for plan mode where you ask questions to the user. how does that tool work". And Claude answered! It has access to a AskUserQuestion tool which it uses for this purpose. What's fascinating about this approach is that the team @AnthropicAI decided that this is the best way this job done. I mean if I wanted my agent to get some info from user, my first instinct would be to add it in the system prompt. Something like "If you feel like you could use user's input on something, ask them clarifying questions instead of guessing". But instead, Anthropic created a tool for agent to user. Here's the description of the AskUserQuestion tool that Claude sees. I'm taking notes as a developer because this is the pattern to use your own AI assistant if you wanna get user input. The goal can be anything - asking clarifying questions, getting feedback, collecting user data. Between this and using tool call to manage a working TodoList, i feel like we have just started scratching the surface of how versatile this tool calling framework is.

I'm SO impressed by Claude Skills. They are the best way to package custom instructions (and code) for the agent to access. (I prefer this to Claude Code's subagents because of the way it handles context) I used the skill-creator skill (comes default inside Claude) to create a skill for designing a presentation like Steve Jobs or Seth Godin, then used that skill to create a PPT about Claude Code. Below is a walkthrough of the results. The output when using the above mentioned skill vs when i don't is so clear. I'm bullish on what this implies for both Claude web and Claude Code. Extremely powerful. @AnthropicAI cooked! 🔥

My sincere petition to @windsurf_ai on behalf of the People to keep choosing simplicity. I don’t want auto-select toggle and a thinking toggle, rules folder, tabs, agent mode, edit mode custom mode or MAX mode. I also don’t want to do a different pricing math for different mode or an optional large content math or a MAX math. Please keep doing what you’re doing and dont go the direction of Cursor. 🙏🏽

Prediction: this new o3 model from @OpenAI is the AI that unlocks the promise of personalised education. It has a never before seen ability to fill in your knowledge gaps and teach you new subjects in an easy to skim but dense tutorial. I just asked it to teach me about Renaissance and Industrial Revolution from the context of @sama’s tweet below.

I always thought that @dvassallo was priced his community's lifetime deal out of the goodness of his heart. I mean why else would you offer lifetime access to - an ever expanding library of exclusive content with the top creators + - access to a super active community that keeps growing for just the price of an annual membership?? He’s basically leaving money on the table, right? WRONG. Today, I learnt that it’s just great business sense on Daniel's part! I was able to ask him these questions on my podcast today. (Look @AndrewWarner i just interviewed another hero) This is the TLDL (too long; didn't listen) of what he said: • Overdelivering is a great business marketing strategy. • Look at Average Revenue Per Customer and you’ll clearly see how this model is superior to any • Churned customers in a community almost always leave with a bitter taste in their mouth. This means, best case scenario, they won’t refer other members to you but worst case - they’ll actively dissuade their friends from joining it. Attaching a quick teaser from our interview below. (If you would like to be notified when the episode drops, sign up for the free email course link in the next tweet)

The funny thing about working @every is that there are 6 engineers in the team and everyone has a different toolkit of working with AI that they absolutely swear by 😂 This is why this team is able to stay at the cutting edge of what's possible with AI. Anyway here's my simple but mighty stack 👇🏽

i shared Claude Code commands that help me on non-technical tasks: - /help-me-market: looks at codebase commits to understand what was shipped and helps me create marketing copy for it (in @ShaanVP's style) - /update-help-center: looks at codebase commits to identify how to surgically update our user facing help center so it's always upto date More about them in the article below

what does your claude code status line look like? i just updated mine to show: - day + what % of day is passed - which Q we're in + how much of that Q is over + week of year - one of the 9 rails doctrines - current branch - tokens used in current claude code session - active model and everything in different colors! 🌈 it was so fun to use /statusline command to brainstorm what's possible and add things to it one by one! @AnthropicAI is the only ai lab that understands "vibes" ❤️

Just found this gem on @aarondfrancis 's blog: "If you don't try hard and you fail, you've got a great excuse! You can say, "I could have succeeded if I had tried! Oh well, I'm not a failure because I know I would crushed it if I had just tried, but I didn't really try, lmao." Hits hard.

my new fav coding setup - @windsurf_ai with sonnet 3.7 thinking mode feels really powerful at agentic tasks. it feels particularly good at rethinking and considering original constraints from the rules and memories after each step.

there's no company more creative than @every! there are so many AI Easter eggs hidden in their entire product. i mean just look at the Teams page - every.to/team this is what they did to my headshot using AI! 😍

idk why no one is talking about it but @windsurf_ai gives you almost 4x more ai usage than @cursor_ai for 25% less money ($15 vs $20) Here's the math: Cursor Pro - 500 fast requests (+ unlimited slow requests which in practice don't work) - each AI message, tool call (edit, analyze, search etc) costs 1 fast request with a premium model like Sonnet 3.7 Windsurf Pro - 500 User Prompt credits + 1500 Flow credits - each message you send to AI consumes 1 user prompt credit and each tool call (edit, analyze, search etc) costs 1 flow credit with a model like Sonnet 3.7 so either Windsurf is bleeding through cash here on their ai bills or they've got some neat tricks under the hood that allows them to offer all this while building a sustainable business.

I've been really impressed with the way Claude Code has started asking questions when you're in Plan mode. So i was curious if i could ask Claude to reveal how this works. I asked Claude - "tell me about the tool you have for plan mode where you ask questions to the user. how does that tool work". And Claude answered! It has access to a AskUserQuestion tool which it uses for this purpose. What's fascinating about this approach is that the team @AnthropicAI decided that this is the best way this job done. I mean if I wanted my agent to get some info from user, my first instinct would be to add it in the system prompt. Something like "If you feel like you could use user's input on something, ask them clarifying questions instead of guessing". But instead, Anthropic created a tool for agent to user. Here's the description of the AskUserQuestion tool that Claude sees. I'm taking notes as a developer because this is the pattern to use your own AI assistant if you wanna get user input. The goal can be anything - asking clarifying questions, getting feedback, collecting user data. Between this and using tool call to manage a working TodoList, i feel like we have just started scratching the surface of how versatile this tool calling framework is.

guys i think i may have created the most amazing Claude subagent 😂 @AnthropicAI is going to the moon now 🚀

I always thought that @dvassallo was priced his community's lifetime deal out of the goodness of his heart. I mean why else would you offer lifetime access to - an ever expanding library of exclusive content with the top creators + - access to a super active community that keeps growing for just the price of an annual membership?? He’s basically leaving money on the table, right? WRONG. Today, I learnt that it’s just great business sense on Daniel's part! I was able to ask him these questions on my podcast today. (Look @AndrewWarner i just interviewed another hero) This is the TLDL (too long; didn't listen) of what he said: • Overdelivering is a great business marketing strategy. • Look at Average Revenue Per Customer and you’ll clearly see how this model is superior to any • Churned customers in a community almost always leave with a bitter taste in their mouth. This means, best case scenario, they won’t refer other members to you but worst case - they’ll actively dissuade their friends from joining it. Attaching a quick teaser from our interview below. (If you would like to be notified when the episode drops, sign up for the free email course link in the next tweet)

My sincere petition to @windsurf_ai on behalf of the People to keep choosing simplicity. I don’t want auto-select toggle and a thinking toggle, rules folder, tabs, agent mode, edit mode custom mode or MAX mode. I also don’t want to do a different pricing math for different mode or an optional large content math or a MAX math. Please keep doing what you’re doing and dont go the direction of Cursor. 🙏🏽

I'm SO impressed by Claude Skills. They are the best way to package custom instructions (and code) for the agent to access. (I prefer this to Claude Code's subagents because of the way it handles context) I used the skill-creator skill (comes default inside Claude) to create a skill for designing a presentation like Steve Jobs or Seth Godin, then used that skill to create a PPT about Claude Code. Below is a walkthrough of the results. The output when using the above mentioned skill vs when i don't is so clear. I'm bullish on what this implies for both Claude web and Claude Code. Extremely powerful. @AnthropicAI cooked! 🔥

Prediction: this new o3 model from @OpenAI is the AI that unlocks the promise of personalised education. It has a never before seen ability to fill in your knowledge gaps and teach you new subjects in an easy to skim but dense tutorial. I just asked it to teach me about Renaissance and Industrial Revolution from the context of @sama’s tweet below.

Claude skills are awesome. in this video, i take a peek under the hood of skill-creator and mcp-builder skills. there's so much to learn here. i believe Skills are the most powerful packaging of customization inside any ai agent. you can literally create a skill with infinite context and it will work because it is designed to use Claude's agentic search capability for information retrieval. the same ability that allows Claude Code to work on big codebases without doing any RAG. i used this to create a skill where i literally put an entire open source repo of a product as reference for Claude and it works! not to mention that mcp-builder skill is probably the best guide ever written on creating high quality MCP servers. if you're into using ai, i highly highly recommend reading through some of the skills in this @AnthropicAI repo: github.com/anthropics/ski…

What's a good @cursor_ai alternative? it has been completely unreliable this week making me realise that i should not be so dependent on one startup for such critical business work.

There has never been a good counterpart to @ycombinator's cofounder matching service - but for indie hackers and people who want to build a lifestyle business. That changes today. Super excited to partner with @nateritter to offer this to the founder community! ❤️ Completely free to join - just come with the right intentions. (DM Nate for the invite link)