A structured training programme introducing practitioners to agentic AI systems, covering context engineering, memory management, agent orchestration tools (such as Roo Code), and practical case studies in project management, data visualisation, and academic research through phased, multi-session instruction.
Semantic Classification
Content
Terminology
Definition
The term agent is contested, and has been for years. The simplest definition is a piece of software that does something on your behalf, using AI.
What are Agents
I have given up following the debate because in a way it doesn’t matter. As a good heuristic Large Language Models are training on huge corpora to predict the next most likely token Chatbots - LLMs that are tuned by Reinforcement Learning for turn based chat Agents Tool use Memory Agency (decision action trees) Minimal oversight Outcome driven Agentic Systems Persistence across sessions Learning (allowed by persistence) Complex tool use Complex tasks Able to create and use tooling Multi-Agent Orchestration Complex organisations of Agentic Systems Long runs times Open ended discovery and knowledge synthesis Systems level problems Expensive, difficult to make safe / secure — working/pages/Agents.md
- Think of agents as sheep dogs. You’re not putting sheep in a pen one by one any more. sometimes the best path to getting the sheep in en masse is a long arc round the field where the agent develops the mass as it goes. Sometimes it’s a structured play with more work up front. This somewhat depends on you.
- My background, why you’re here, what you can expect
- course structure (multi session, not completely interdependent, simple pricing)
- background and rationalle
- chatbots and the agent / agentic boundary are confusing
- you probably already use them (deep research)
- compare gemini pro deep research with roo (first view of it)
- why this complex way
- talking about memory
- consistency, flexibility, scale
- data safety and privacy
- longer run times, more thinking, as much as you need
- price / performance
- power use and the start of vibe coding
- you probably already use them (deep research)
- phased approach
- Phase one (today)
- introduction to vscode
- introduction to roo code
- context management and context engineering
- demonstrate agents, see where we’re going (visionflow)
- show setup of the system (link to online homework)
- show how to find all this guidance
- gemini (free) (more like agents)
- claude code (paid) (more like agentic)
- case studies using agentic claude
- -Project management (120 kids doing their own thing)
- -Data visualisation (new python tools)
- -Academic research and verifying sources. (latex)
- phase two (tomorrow)
- Mcp security course. Context, sheaf of papers
- use the search on agentics video site Agentics Foundation Video Portal