Hey guys, I need help, advice. I'm building an automation for a client to create automated quotations when he receives email requests. This sounds pretty straightforward, but it's not. There's one source of emails that comes always from the same sender, and that one is easy; it's forwarded from a factory form. But then I want the automation to scan the inbox every 30 minutes to check if there are any other quotation requests. The inbox is scanned, and labels are attributed to the emails, classifying them, After that, an automation runs and creates the draft. But the draft is complex because it has to go through 500+ pages of PDF, which is very complex. So it has the PDF split into categories and sends the draft only with the pages of that specific product. There are some products that it actually gives the final price for because they don't have combinations of fabrics or colors. So it is pretty complex. My question is, I build this with Google Workspace features like buckets and stuff like that for the "rag" and I have a massive complex make.com automation that you can see in the image, Is this the best way to do it? for real? with such complex make.com automation? Can't I just tell claude to do everything. I asked Claude and COdex and both guide me through this. It's working but it's very complex and I was wondering if it could be simpler... Sorry for the essay... Thank you

Hi everyone, I'm Tejas Hade here's where I'm stuck: I followed the 6-hour NetHacks YouTube video and got excited with too many ideas. Instead of starting small, I tried to break into all the skills at once. Now I have a messy AIOS with half-finished skills, lost context, and therapy transcripts that burn through my tokens. I'm completely stuck and don't know whether to start fresh or salvage what's there. I am also a member hoping to reach Level 3 in the free community to access the AIOS course. I'd really appreciate a few likes/comments to help me get there ⚡️ Any quick wisdom from those who've been through this chaos would be a huge help. Thanks in advance!

SO part 1 week i have replaced my whole team with Ai agents, all are claude code basically. I have 1 main manager (claude code) which manages Team under it: Ads lunching (Media Buyer) Store Developer Creatives Maker and few more... But I see the manager is missing a lot of shit even with solid SOP's I ask it to check every details, and it lacks a lot lately. Yes I am using Obsidian and everything is getting notes saved and also claude itself clears so there is also no context bloat issue as of I know. But the mistakes it is doing is: It launched some campaigns on different meta page which i clearly mentioned earlier in the SOP that ads launcher needs to launch with this page. But manager checks just on top top things, doesn't go deep inside. I would be really happy if someone can help me with this or in same shoe and fixed these problems. Like I try my best the all agents execute properly without any dumb issues, but they kept doing. I am hammering my head right now tbh. Any help will be appreciated! @Nate Herk

Past decisions becoming an INVISIBLE BIAS. An agent solved a problem one way last month. Now it keeps repeating that approach, not because it’s still best, but because its own history makes it feel familiar. That’s a subtle failure mode: precedent quietly becomes policy. Past decisions can be useful context. They aren’t proof that the same decision is right today. A practical safeguard: 🦉 Retrieve relevant past decisions. 🦉 Revalidate today’s facts, constraints, and system state. 🦉 Compare the old approach with viable alternatives. 🦉 Override history when current evidence disagrees. One rule worth testing: memory should inform the decision, not pre-approve it. Even an AI can develop “we’ve always done it this way” syndrome. How would you stop an agent from reinforcing its own past mistakes without making it re-investigate everything from scratch?

I’m setting up an AI OS. There’s lots of material on building one and cloning your context from ChatGPT to Claude, but how are people keeping them synced afterwards? If I update a project, make a decision or change an instruction in one, I’d like the other to pick that up without manually copying everything across. I found a few relevant discussions here: - ChatGPT update thread: Israel Yanez described using a shared workspace and an “archivist” skill to save sessions to local logs and SuperMemory. - Unlimited Memory for AI: keeping context, tasks and insights in shared Markdown files. - Hermes/Claude sync thread: keeping one main store for live memory, with GitHub used for versioning and backups. Has anyone got this working reliably between ChatGPT/Codex and Claude? What’s your setup, what updates automatically, and what still needs a manual handover? Would appreciate a practical walkthrough or template—especially one that works on Windows. Or am I overthinking it. I am only beginning to build my first AI OS now. Is the idea to build it, clone it, and then not worry too much about real time syncing?