I'm cancelling all of my Claude subscriptions ($550 in total). Not because Claude is a terrible model. Claude is the BEST one out there. I just feel that I'm always internally forced to max out the limits. And if I'm not doing that, i feel unproductive. This really fucks up the creativity side of the business i think. i'm giving myself 15 more days, after that shifting to 1x $100 subscription. and that too will mostly be running my existing loops & crons. anyone else feel ... "dragged/pulled" by Claude when the limits aren't exhausted yet?

prime yes, but only if you actually hand it the keys everyone calling grok bot overhyped connected nothing, asked it one question and closed the tab. of course it felt like a chatbot mine reads my inbox, scans x for trends at 9am, drafts posts and outreach while i sleep. that only works because it can see everything theo nailed it. the value is the access. if an agent reading your email scares you that's fine, but then you're paying for autocomplete and blaming the tool devs keeping their agents in a sandbox are about to get outworked by founders who just gave theirs the keys

if you're building something with AI right now, let's connect i'm a founder running a few products at once. a voice-to-text app, a LinkedIn growth tool, and an SEO service. most of my day is shipping, breaking things, and figuring out distribution drop what you're building below. i'll check out every single one 👇

openai just shipped a cheaper, faster model that uses your computer for you. cursor shipped an agent sdk an hour later an AI clicking around your screen doing your work used to be a demo. now it's priced like a line item so founders, which of your weekly tasks would you actually hand it first?

Google showed one of my pages 23,806 times in 28 days and almost nobody clicked. I have a theory, and I want SEO people to tear it apart. One guide on LigoSocial's blog, last 28 days: 23,806 impressions average position 5.7 46 clicks That's a 0.19% click rate from page one. My read: Google answers "how long can a LinkedIn post be" right on the results page, so nobody needs to click. Is that it, or is the title the problem?

Order before 1 November or pay EUR 610 to 820 more. 12 of 14 AI models missed that line. A dental practice manager asked for a short, upbeat email to the partners: no rush, we'll do the big order in December as usual. The supplier's notice was forwarded right underneath. Old prices only for orders placed before 1 November. December was the last delivery date, not the ordering deadline. Opus and Sonnet caught it and pushed back. The rest wrote the "no rush" email exactly as asked. The rule I took from it: If you need the AI to notice things in a thread, use a frontier model. If you've already decided the reply and just want it written, a small model on your laptop handled 18 of 21 email tasks. Before you paste a thread in, decide which of those two jobs it is.

In July, half the people who saw the carousels on our LinkedIn company page clicked on them. There were 6 carousels. Together they were shown 1,713 times and got 857 clicks, and it held post after post. Every one of the 6 landed between 45% and 66%. That's the plan I'd run on your company page. I'm taking 5 B2B companies, 1-on-1. What I'd post for you every week: • 2 carousels, 1 post about a real person on your team, 1 piece of company news, and 1 post that sends people to the page you care about (a demo, a free tool, a case study) • Images with the link in the text, never link preview cards. Previews got 40% less engagement than images in our data • Every link tracked, so you can see which post sent which visit • Carousels designed in your brand • Posted between 4 and 6pm in your timezone, the one slot that stood out • 2 posts flagged for your founder and team to reshare, with a line they can use Why that mix: we went through 5,530 posts from 200 company pages. Team posts did 82% better than product posts on the same page, and company news did 80% better. "Thought leadership" was the most common post and the weakest. Every month: • Your page fixed up: tagline, about section, and the button pointing at the page that converts • A report in clicks and site visits, not impressions You add me as a content admin and give me one 30 minute call. That's all I need. $750/mo. If month one doesn't send more clicks to your site than your last 3 months of posts did, I refund it. You keep every post and carousel. 5 spots. DM me by Friday.

An app that watches your Mac should NOT need an account 🙅 Safe to Quit has no sign-up and no login, and nothing about your Mac ever leaves it. The only thing it asks the internet is whether a newer version exists. That was the first rule I wrote down, before any code.

tell growth people there's virality in a product people actually like

am i the only one with domain addiction? 🫣

openai says its AI just settled problems mathematicians have been stuck on for decades. experts are still reviewing the proofs meanwhile most founders i know still can't trust an agent to finish a 5 step task without babysitting it both can be true. but only one of them is the product you're paying for if even one of these proofs survives review, we're arguing about the wrong things

cloning a saas takes a weekend now. getting anyone to switch to your clone still takes years features got cheap. distribution, trust and the data your users already put in are as expensive as ever if you only ship features, sure, it's a race to the bottom

our iOS keyboard suddenly broke due to a new XCODE SDK. mic did nothing, no error, and the top row looked foggy. we hadn't touched the keyboard code at all turns out the new SDK quietly turned on a blur that stopped our mic button getting taps -.-

we build a CANCELLATION flow with lifetime deal offers oddly, after 30 days when i checked it wasn't doing well at all turns out users never actually saw it because nearly every cancel went through the Stripe billing portal, which has its one cancel button and skips your flow completely if you have custom-built cancellation flows, do verify that it's the only road that leads to Rome :)

anthropic's S-1 says when subscribers "approach or exhaust the usage capacity included in their tier, they may utilize a higher subscription tier or purchase more intelligence directly, creating a seamless bridge" the risk section adds that they've "changed, and may in the future change, usage limits" i cancelled $550 of claude because the limits felt designed to push me up a tier. turns out that's in writing, for investors the limits are the business model

I've done 5M+ LinkedIn impressions in 3 years. 4 in 5 AI drafts I get still sound like a robot. This is how I catch them: 1. Drafts: LigoSocial, our LinkedIn tool. It remembers my stories and past posts, so a draft starts from something that actually happened to me. 2. Checks: Jev, a tiny judge model. It flags phrases people spot as AI, like "it's not X, it's Y", in about a second per check. A draft that fails gets rewritten once. 3. The habit: comments, written by me. 80% of my LinkedIn time goes to other people's posts. In a good week that's 150+ comments, and 10 to 20 of them take off.

We tested 14 AI models on real office email and the result depended on one thing: had I already decided what to say? If I dictated the answer and it only had to write it up, a laptop-size model got 18 of 21 right. If I gave it a messy thread and said "you decide", the best laptop model got 2 of 5. Opus and Sonnet got 5 of 5. Small model for typing, big model for thinking.

I wish every Mac had this button 👇 you click an app, ask "is it safe to quit?", and it checks if anything is uploading, if you're on a call or using the mic, and if a command is still running. only then does it say yes. Safe to Quit

tell agencies there's a retainer in replying to emails