Aaron Makelky

people have no clue how much better @comet browser is now that Claude Opus 4.5 is the default model I just watched it work on a complex task for 20 minutes straight, executing 185 steps from 1 prompt

I used to scroll YouTube every morning looking for something worth watching while I workout. Sometimes I'd find something good but usually I'd waste 10 minutes picking, then half-watch a mediocre video. So I gave my agent a skill that doom scrolls YouTube for me. It uses the YouTube API to search for videos based on my projects and calendar. Each video gets its own page in a @NotionHQ database with a learning exercise that connects the content to whatever I'm actually working on. And it's ready every morning before I go down to the gym. An example of how powerful this is: I'm speaking at a conference this week, so it started pulling public speaking videos. It knows I've been deep in @claudeai Cowork, so those topics have shown up too. All from my project file and calendar. I didn't tell it to do any of that. This took about 30 min to set up, and has already made my workouts more efficient and the videos I watch far more actionable.

@huynhquoctoan9x @comet I haven't used Safari, but Comet blocks ads as well as anything I've ever used

@theaaron @comet Nó có chặn quảng cáo tốt như trên Safari không?

Happy “best browser is now on iOS” Day to all who celebrate @comet sync was complicated to set up, but works great between desktop -> mobile

If you are a @ChatGPT refugee exploring @claudeai This is what I wish someone would have told me when I first set up Claude Cowork: I would load up a session of Claude Cowork and immediately start prompting like I’m still in a chat app. Ask a question, get an answer, close the tab, start over next time. That works fine. But it's not how Cowork is designed to be used. The difference with Cowork is that it maintains context across sessions AND it can work locally with your files. But you have to give it context first. Three files make the difference: about-me.md: who you are, what you do, your constraints working-preferences.md: how you want Claude to work with you voice-and-tone.md: how you actually sound when you write Cowork reads the relevant context before every task which means I don't need to repeat myself or prompt from scratch. The setup takes 10-15 minutes if you do it manually. Or you can use this: Guided setup (free): github.com/a-makelky/clau… Paste the prompt, answer the questions, get your context files generated automatically. Drop them into Cowork and you're done. I'm not sharing my actual context files (they're too specific to my work to be valuable to others anyways). But this setup creates personalized workspaces for whatever you're doing.

@denisyarats @comet Your DMs are closed but yes I’d love to share

@theaaron @comet could you dm why it was complicated?

@nbaschez Or a /command works great too

This is good The one thing I'd add is that neither Claude nor Codex are amazing at just magically using the skill you want them to Solution is to put a short keyword (eg "use SRE") in the description, so that you can easily append to your prompts

@x6c75 Did your @openclaw wish you happy birthday?

its my birthday, i feel OLD

it's beautiful...

go go @claudeai as your video team assistant anyone with a @descript account can create an API key and pass your video & audio projects to Descript through the API, including via @zapier, MCP, and CLI DM me if you need help or have feedback for the team @elrolio @burkenstocks

@tom_doerr Used Claude to help people who wrote with Claude to not sound like Claude…got it!

Removes AI writing patterns from prose github.com/hardikpandya/s…

@samuelchenard @lobster_mail first thought: only service I've seen where the agent can actually create its own email, the others require me to auth and create, then pass info to the agent

@theaaron @lobster_mail @theaaron two birds one stone, we verify you're not a bot and get a little visibility. hope it's useful and let me know if you have any feedback!

@samuelchenard free marketing as verification, I like it! @lobster_mail account ID: acct_0-nup1oSrBxp1fwU Verifying my AI agent's @lobster_mail inbox acct_0-nup1oSrBxp1fwU

@MatthewBerman @Zai_org GLM 5

What’s the best open source model for coding right now? Any size 😆

proud father moment: my son vibe coded a math review game that has a tank battle mini game once the user answers enough questions correctly his school filters many game sites, but we moved his app to his own domain and now he is having his classmates play it at school "because they need math practice too dad" clutch @get_mocha project @nichochar 🫡

Codex is great for everything but design. I had Opus 4.6 create design specs, then Codex built something that looked great. All of them are getting so easy to use, Replit & Lovable are simple. Claude Code or Codex with GPT 5.4 if the user knows how to move files around and deploy

@theaaron @comet What do you recommend for creating websites right now with AI?

can @comet redeploy a website if you give it instructions and a tab group of sites needed? Opus 4.6 Thinking...

@collin_ruth89 @channel It walks you through creating personalized files for your coding agent to work in so it's not just the same thing as Claude generically.

@theaaron @channel Can you explain to me the point of this like I'm a 5th grader?

This morning I gave every FounderBrands employee a credit card to spend on AI. Here's the message they received in Slack: Hey @channel I’ve set up a virtual credit card with $300/month dedicated to one thing: finding AI tools that make your work better. Everyone in this channel has spending authority on this card. — What we’re looking for: Tools that make your job easier, increase your capacity, or automate the repetitive stuff pulling you away from real work. Where AI wins: - Research - First drafts and ideation - Data analysis and pattern recognition - Repetitive processes that drain energy - Pulling together information from multiple sources Where you win: - Strategy and planning - Creative problem-solving - Building real relationships - Final judgment and quality control The goal is to let AI handle the former so you can spend more time on the latter. Two rules: 1. Over $50/month on a single tool? Ask me first. Under $50? You have full authority. 2. If something isn’t worth it, cancel it. One last thing: Every FounderBrands employee gets compensated based on the volume of work you can handle. AI done right means you can handle more, which means you earn more. Start with your biggest, most draining problems and go from there. --- Follow me @collin_ruth89 for updates on how this goes.

this is hard to do @OpenAIDevs

@IMJustinBrooke no amount of pre-emptive prompting or instructions can keep them out, but run this after and it will: github.com/a-makelky/ai-w…

SERIOUSLY... who trained the em dash into AI? I want his name. Bro... TAKE IT OUT! I have 9 different rules in my account banning em dashes and they still slip in — somehow.

@collin_ruth89 @channel Smart. Claude Max=$200 Would love to see a follow up on what they spend it on and the results you see!