Part 1 · Overview
What we’ll cover
The stack people are actually shipping—not slide decks.
- How agents really work (beyond the hype)
- Building the core loop: tools, planning, and memory
- Using Claude Code to create real task execution workflows
- Extending into OpenClaw to operate real software
- Turning “AI assistant” into something that completes jobs end-to-end
Part 2 · Framing
Why this matters
Most “agent” demos stop at chat. This bootcamp is about systems that keep state, take action, and run on your machine—something a browser tab alone usually can’t do.
1 · Thesis
Most “AI agents” today are just scripts with better marketing.
A real AI employee has context, uses tools, executes tasks end-to-end, and improves over time. That’s what we’re building here.
2 · Definition
What a real AI employee does
Capabilities that separate employees from chatbots—and from a browser tab that only sees the current session. A tab in Chrome can answer questions; a local employee is built to live with your work.
- Tools and real actions Calls APIs, browsers, and your stack—not just text. Email, social, calendar—integrations, not copy-paste.
- Context and persistent memory Remembers what matters across sessions and tasks. Preferences, projects, and contacts—not just the last prompt.
- Long-running and end-to-end Sticks with hour-long tasks—research, builds, batch work—until real outcomes ship, not half-finished drafts or a single reply and stop.
- Improves Gets sharper as you tighten loops and feedback.
- Always available 24/7 on your machine—no cloud tab timeout.
- Private Your data stays on your machine, not someone else’s servers.
Part 3 · Program
Program: three parts
Agent fundamentals, Claude Cowork in practice, then OpenClaw setup and operations—end to end.
1. Agent fundamentals
- s01–s02: Agent loop + tools — the core abstraction (read → think → act → repeat)
- s03: TodoWrite — planning before acting
- s04–s05: Subagents + skills — modular intelligence
- s06: Context compression — scale beyond context limits
- s07–s08: Tasks + background execution — real-world workflows
- s09–s10: Agent teams + protocols — multi-agent coordination
- s11: Autonomous agents — self-directed execution
- s12: Worktree + isolation — production-grade reliability
2. Claude Cowork
-
1. Mindset shift
From “chatbot” → “AI coworker that does tasks”
-
2. What Cowork is
- • Lives on your computer
- • Works on your files (reads, writes, builds)
-
3. Core setup
Folder = AI brain
- Context (docs) + Data + Tasks
-
4. How to use it
- • Give tasks, not prompts
- • Iterate: assign → review → refine
-
5. Key use cases
- • Data analysis (CSV → insights)
- • Content creation (posts, reports)
- • Ops automation (docs, workflows)
-
6. Advanced
- • Save tasks as reusable “skills”
- • Multi-step workflows
- • Background execution
- • Use your computer: read your screen and click
- • Claude dispatch: use your phone to work
3. OpenClaw
Hands-on setup and operations—channels, memory, automation, and security.
- Complete setup from scratch, including all dependencies
- Connecting WhatsApp, Telegram, Discord, iMessage, and email
- Configuring persistent memory so your AI learns and remembers
- Posting to social media (with platform-specific formatting gotchas)
- Setting up automated tasks with cron jobs
- Security configuration to control who can access your agent
- Running the gateway as a background service