This video introduces how to use OpenClaw (an open-source framework) to build your own AI Agent team and achieve 24/7 automated operations. The speaker shared how he replaced human assistants with four AI employees running a $600 Mac Mini. Introduction to OpenClaw AI Agents Core architecture: 4 agents 1. Max Chief of Staff -Daily communication through Telegram -Manage the other three agents -As COO, collect all results and report them 2. Sage - Content Specialist -Draft tweets and analyze performance -Research competitors and discover hot topics -Generate 5-7 creative drafts of tweets per day 3. Nos - Trading Operations Agent -24/7 monitoring of Polymarket trading robots -Check for errors and investigate losses -Report health status every two hours 4. Nova - YouTube Strategist -Track data analysis performance -Research hot topics -Propose video ideas and script suggestions Example of Daily Workflow The briefing received automatically at 7:30 AM in the morning: -Sage has captured the content of competitors' tweets -Based on actual performance, we identified hot topics and drafted 5 creative tweets -Filter out the last 107 rejected drafts to avoid duplication -Key: This is a fully automated task, no need to actively inquire Workflow: -View all drafts through MaxQ (a tool built with AI Agent assistance) -You can directly approve or reject and provide reasons -AI will continue to learn based on each rejection feedback Transaction monitoring of Nos: -Polymarket cryptocurrency prediction market trading robot -Over 1500 trades with a win rate of 92% -Automatically identify the root cause and fix or report any issues that arise Nova's video management: -Creative approval → Script generation (using your voice) → Editing → Shooting plan -Track metrics after uploading, analyze videos with good/bad performance -Adjust the subsequent content strategy ️ 5 Steps to Build a Framework Step 1: Track for a week Core findings: -70% of work time is spent on repetitive and monitoring tasks (checking and analyzing, scanning Twitter, drafting initial drafts, watching trading bots) -30% is dedicated to strategy, customer communication, and final creative decision-making Step 2: Start with an Agent -On the first day, only Max (Telegram universal assistant) was established -Two weeks later, I found myself repeatedly asking the same question ("What's trending on Twitter? What can I talk about? ) -Then establish Sage (dedicated content agent) -Key point: Capture the work you do yourself, and the next agent should handle these tasks Step 3: Give your Agent a personality file OpenClaw is commonly referred to as` http://soul.md `： Have opinions Be concise Call me out if I'm about to do something dumb -Agents without personality files sound like other chatbot wrappers -General and vague, unlike the people you want to work with -And also` http://user.md `Tell it everything about you: business, goals, preferences Step 4: Establish a feedback loop This is key: most people give up on setting up AI agents and say 'AI is not ready yet' when they get general results. -Every time Sage sends a tweet idea, either approve or reject it and provide reasons -Generic (posted last week) - No one cares about this -After 100+rejections, Sage reads all rejection records before writing new content -Established an automated system to classify each reason for rejection and track topics of boredom -Important: Place the feedback loop at the core of the workflow -The first month will be difficult, but in the third month, you will be able to write the draft you want Step 5: Automated scheduling instead of just tasks Level One - Responsive Agent (responds when asked) Level Two - Active Scheduling Agent that also works while sleeping Key: Set once and run every day. When you wake up, the work is done. cost analysis Monthly expenditure: approximately $140/month Project cost API call (four agents)~$100/month Mac Mini computing cost~$20/month (running 24/7) OpenClaw $0 (open source and free) Other APIs (Perplexity, Google Trends). ~$15-20/month One time investment: Mac Mini=$600, recouped in the first month Compared to human assistants -AI Agent: $140/month, working 24/7 -Human assistant: at least $5000/month, only 8 hours/day >The initial cost was $300/month (due to the use of expensive models to handle simple tasks), which was reduced to $140/month after optimization: using inexpensive models for simple tasks and intelligent models for complex reasoning. Efficiency improvement: Output increase: 3 times Reduced working hours: only 1/4 of the original Time savings: reduced from 8 hours to 2 hours per day Explanation of the Current Situation AI agents are not magic: -Sage still writes rejected tweet ideas -Nova sometimes cannot identify the cause of trading errors -50% of Nova's recommended ideas are still rejected Shift from completing all tasks on your own to reviewing and directing completed work. The nature of work has shifted from execution to management and operation. This is a real-life case: the speaker built their own AI team through OpenClaw and achieved 24/7 automated operations. The key is to gradually establish, provide personalized files, establish feedback loops, and ultimately achieve schedule automation