Hands-on with Agentic AI

If you’re here from the lecture, welcome! Add me on LinkedIn.

If you haven’t already, please sign up for GitHub and GitHub Copilot Free — that’s free access to the agent and models we’ll be working with.

During the lecture, we’ll be deploying AI agents for:

1. a simple data extraction task, and
2. Deep Crew, a classroom tournament where your robot plays every other robot in the room.

---

The slides and source code are available under CC BY-SA 4.0 license.

Reading List

When writing prompts, I recommend keeping up-to-date on industry best-practices. Some good places to start:

• Claude Prompt Engineering Docs
• OpenAI Prompt Engineering Guide
• PromptingGuide.AI

If you’re looking for skills, start here:

• The anthropics/skills collection
• The openclaw/skills collection
• https://github.com/blader/humanizer

There are plenty of skills for specific tasks and tools. If you can’t find exactly what you need, you can always deploy the Skill Creator skill. Feed it some documentation and examples and it will generate a skill for you. It may be the most important skill of all.

(Be warned: its easy to smuggle malicious code and prompts in with skills. Check them carefully.)

For more general reading:

• My AI Platforms Lecture, including open-source as a defensive and offensive maneuver
• My Early Stage AI Startups Lecture
• This CGP Grey video
• Sizes of LLMs
• Chesterton’s Fence, or in short: “Do not remove a fence until you know why it was put up in the first place.”
• The Men Who Killed Google, for how overreliance on metrics (rather than product quality) leads to short-term gains at the cost of long-term goals.
• Apodex 1.0, a bet on swarms of small models that verify and rewrite each other’s work — a possible future alternative to scaling one giant model.

As always, turn your critical thinking skills on and carefully engage with the claims each source makes. Test what you can