Unpopular opinion: big skill packs make Claude Code worse for client work. Every extra skill is another chance for the wrong one to trigger, and none of them know which client you're working on. So I went the other way and kept only the 5 procedures I actually use every week, and open-sourced them: 1. client-workspace: one folder per client with its own rules, voice and approved claims, so Client A's tone never ends up in Client B's email 2. scope-guard: checks every "can you just quickly..." against the signed scope, quotes the extra, and drafts a polite reply (offers a paid add-on or a swap) 3. quick-quote: prices a project from 6 numbers about the client's workflow instead of your hours, and tells you when you're underquoting 4. offer-builder: turns "I do AI automation" into a fixed-scope offer with a price ladder 5. cold-email-agency: 4-step outreach sequences for small business owners There's also a full worked example (a fictional plumbing company) so you can see a real scope check end to end: 3 asks in one client email, one in scope, two quoted, reply drafted. MIT licensed, no email needed: https://github.com/CalebAnderson-Coder/ai-agency-os-starter If you try it, I'd like to know what breaks or what you'd cut.

Here’s what I worked on: - Added and imported contacts - Explored pipelines and lead management - Learned how to move leads between different pipeline stages - Created a workflow that automatically adds a lead to the pipeline when someone submits a form The most interesting part was connecting the form submission to the pipeline through automation. Instead of manually adding every new lead, the workflow handles that step automatically. I’m learning that the real value of automation isn’t just saving time it’s building systems that make everyday business processes simpler and more organized. One step at a time, one workflow at a time! 💡 #GoHighLevel #GHL #WorkflowAutomation #CRM #LeadManagement #LearningByDoing

Been building this for a while and it's finally out. It's a Claude Code setup for running several agency clients without their stuff bleeding into each other. What's in it: - a separate workspace per client (voice, claims, scope) - 14 procedures and 5 agents - a reviewer agent that checks the work before it reaches you - automation that only writes drafts, so nothing goes out without you approving it Honest note: so far I've only tested the skills on fictional client setups. If you run it on a real client I'd like to hear what breaks. There's a free starter if you want to try it first. Link's in the comments.

been testing the new full duplex speech to speech model on my AI receptionist and honestly kinda shook short here: https://youtube.com/shorts/6wk7j70dWVg?si=Bc9llSBp8XIwNOGd so what’s different? old setup (cascade) was 3 models stitched together: speech to text, then an LLM, then text to speech. every hop adds lag, and it has to guess when you stopped talking. that’s why voice bots either talk over you or leave dead air. next step was turn-based speech to speech. one model, audio in, audio out. way more natural and faster. but it’s still walkie talkie. it waits for you to finish, then it talks. interrupt it and it gets confused. full duplex listens while it’s talking. so you can cut in, say “wait actually”, or just go “mhm” and it adjusts on the spot. no guessing when your turn is over. it just feels like a convo. for a receptionist that’s the whole game. real callers interrupt, ramble, and change their mind halfway through a sentence. anyone else messing with this? curious what you’re building 👇

Hey everyone, Adwait here. Almost exactly a year ago, I made a post in this community. I was preparing to leave India for the UK, finishing the next chapter of my education, and trying to build something called Cornerman AI. I had an idea, a laptop, a fascination with automation, and absolutely no clue how much I still had to learn. Then life happened. I moved to London, finished my university journey, met people from completely different backgrounds, trained, explored, failed at things, got better at others, and started seeing business and life through a different lens. One thing I couldn't do while studying in the UK was actively run my business under the restrictions of my student visa. So Cornerman AI had to take a back seat. But the idea never really left. Fast forward to October 2026. I'm 21, back in India, living in Mumbai, and I've decided to give this thing a real shot. Not as a side project. Not as something I'll get around to someday. But as a company I genuinely want to build. Meet Cornerman AI. 🥊 We're focused on helping gyms, martial arts academies, and fitness businesses operate more efficiently using AI and automation. The idea is simple. A great coach should spend more time coaching. A gym owner should spend more time growing their business. And neither should have to spend their evenings chasing leads, manually following up with members, or juggling endless administrative tasks. That's where we want to come in. Over the past year, I've had the opportunity to speak with gym owners, experiment with different automation systems, and work on a few real-world pilots. And something became very clear to me. The hardest part isn't building an automation. It's understanding the human problem well enough to know what deserves to be automated. That's something I'm still learning. Right now, I'm building Cornerman AI from Mumbai. Some days are exciting. Some days I'm questioning everything. There are technical challenges, sales challenges, moments where things click, and moments where I realize I have to go back and learn something from scratch.