I just put 1 message and Claude hit the usage Limit 🤣🤣

Hey all, I'm looking to start a weekly Q&A that will answer all of your questions asked directly from here! How it'll work... Every week, whatever burning questions you have, I will review all of the questions from the post made inside of this group and answer them live on a date to be determined! The stream will be recorded and uploaded here as well for your review later... If you have any other ideas on how to improve this to ensure it's as beneficial for all of you I'm all ears, but I'd love to hear your thoughts on this so drop those down below! :)

Anthropic doubled Claude Cowork limits for 30 days. Most people will use it for small tasks. I'm using it for the workflows I've been avoiding. The ones I thought would break halfway. Here's the real story: This is a price war between labs. OpenAI uses fewer tokens. Anthropic responds with more room. When two giants fight - we win. My plan this month: Run my most complex workflow end-to-end. If it works -I've found something to systematise. What task would you run if limits weren't a problem? 👇

24 hours in, real money on the line, and I've barely done anything else. This isn't the launch-day hype video — it's what actually held up after a full day of building with Fable 5, the one shift that changes how you use it, and the stuff that'll annoy you. - The mindset flips from task to responsibility. Every model up to Opus had you babysitting: give a task, then go check it. With Fable that checking instinct mostly goes away. Anthropic's own example: stop saying "look at this crash report" — make it responsible for keeping the app from crashing. Karpathy called it a step-change ("software on a tap"), though fair warning, he just joined Anthropic. - Three prompting rules that moved the needle. Point it at your hardest problem (easy stuff undersells it). Let it interview you before it builds (single biggest upgrade to my results). And tell it why — who it's for, what it feeds into. - Two real builds, one prompt each. A local-first business hub web app that ran unattended for ~41 minutes, tested its own server end to end, and even fired a real Mac notification (one tab shipped broken, being honest). And a 3D CAD editor it named "Forge" that then designed a watertight, print-ready phone stand from scratch — adding a cable slot I never asked for. - It only pulls ahead on hard work. One-liners feel like Opus. By the end of the day the bottleneck had flipped: it stopped being the model and became how clearly I can describe what I want. - The annoying parts are real. ~2x Opus usage (one build ate 119k tokens in a single turn), guardrails so jumpy they tripped on a phone stand and silently flipped me back to Opus mid-build, and it's slow — hard requests run for minutes. - My take: specialist, not a daily driver. Point it at your hardest, longest, most-expensive-to-get-wrong work and it pays for itself in one good run. Everything else, keep your cheaper models. 📎 Full 24-hour field report PDF pinned below — the exact prompts, the self-QA checklist it ran, and the 3 downsides to plan around.

They have a follow-up problem. I was mapping a simple real estate workflow recently and noticed something interesting. A buyer submits an inquiry. The sales agent is on a property visit. Another team member is already busy. The lead waits. By the time someone follows up, the buyer has already spoken to another agency. If I were redesigning that process, I'd structure it like this: Instant acknowledgment after every inquiry Automatic CRM update Immediate notification to the assigned agent Automatic reassignment if no action is taken Optional appointment booking built into the workflow To me, this isn't really an AI problem. It's a workflow design problem. AI and automation are simply tools that make a good process run consistently. Curious to hear from agency owners or automation builders here: What's the biggest bottleneck you've noticed in lead follow-up workflows?