Token management is real. I've seen people burn $10 in 2 minutes without realizing it. Here's the framework: → NEVER use your expensive model (Opus) for heartbeat/monitoring. Use Gemini Flash or a cheap local model. That alone can save you hundreds per month. → Use sub-agents for simple tasks. Your main brain should delegate, not do everything itself. → Watch your context size. The bigger the context, the more tokens every single message costs. Keep it lean. → Set up token tracking. Run /status regularly. Know what you're spending. → Don't paste your entire chat history into every prompt. Be surgical with context. Realistic budget: If you're smart about routing, you can run a solid AI agent setup for $100-200/month on API. If you're sloppy, you'll hit $500+ easy. The golden rule: if it makes money, there is no budget. Do more of it. But TRACK what's making money vs what's just cool. Don't trip over dollars to pick up pennies.

Both pdfs have been added to the Classroom

Yes. But I won't sugarcoat it — the first 2-4 days are painful. Here's the reality: → The install itself isn't the hard part. Following a YouTube tutorial gets you running. → The REAL work is configuring your core files and teaching the agent what to do. → 99% of YouTube OpenClaw videos are pure hype. The flashy demos skip the messy setup. → When something breaks (and it will), paste the error into Claude or ChatGPT and ask it to fix it. This works 100% of the time. Best advice I got: treat AI as intelligence, not a tool. Don't give it step-by-step instructions. Give it a GOAL and work backwards together. Ask it questions. See how it sees the problem. If it sees it wrong, give it more context. The people crushing it right now aren't the most technical. They're the ones who put in 10-14 hours a day for a few weeks and rebuilt everything 3-4 times. There's no shortcut. But the compounding returns are insane once it clicks. If it was easy, it wouldn't be that great of an opportunity.

This is the #1 problem everyone hits. Here's what actually works: → Your core files (MEMORY.md, SOUL.md, IDENTITY.md) need to be TIGHT. Under 200 lines for memory. If those files are messy, everything downstream gets messy fast. → Lower your context window. Don't let it balloon to max. 200K is plenty. When it compacts at 50-60K, you lose quality. → Reduce how many past messages it loads. Default in Slack is 30 — I dropped mine to 10. Most tasks complete within 10 messages. → Have it SAVE work before it compacts. This is huge. Train it to write progress to a file before the context gets wiped. Think of it like this: your agent's memory is a whiteboard. A small, clean whiteboard beats a massive cluttered one every time. The people who get frustrated with AI agents almost always have a file problem, not a model problem. Either you need a file, it's in the wrong place, or the info is in the wrong file.

It's not either/or. They do different things. → Claude Code = best for BUILDING things. Writing code, creating tools, setting up integrations. → n8n/Zapier = good for simple triggers and workflows. Visual, easy to maintain. → OpenClaw = the BRAIN that ties everything together. Memory, context, personality, 24/7 autonomous operation. Here's the thing most people miss: OpenClaw can USE Claude Code. It's not a replacement — it's the orchestrator. My agent uses Claude Code as a sub-agent for coding tasks. It delegates. Where OpenClaw wins is memory and continuity. You can say "based on everything we've discussed about our TikTok strategy over the last 3 months, what should we prioritize next?" You can't do that with Claude Code. It doesn't remember yesterday. Start with Claude Code if you're brand new. Build useful things. Then graduate to OpenClaw when you need an always-on system that learns and remembers.