Creating agents
An agent is an AI system that works through steps on its own: it reads inputs, decides what to do next, calls tools and returns a result. This course moves from no-code builders to production patterns.
What you’ll learn.
Agent fundamentals
The agent loop: model, system instructions, tools, memory and stop conditions. When a fixed workflow beats an autonomous agent.
No-code builders
Copilot Studio, custom GPTs and Claude projects and skills. What they handle well and where they hit limits.
Tool use
Function calling with JSON schemas, designing tools a model can use reliably, and the Model Context Protocol (MCP) for connecting agents to internal systems.
Retrieval (RAG)
Grounding answers in your documents: chunking, embeddings, hybrid search, citations, and keeping document permissions intact.
Multi-step and multi-agent
Planner and executor patterns, hand-offs between agents, and human approval steps before any action with real-world effect.
Evaluation
Test sets, success criteria, regression tests and step-by-step tracing, so you know an agent works before staff rely on it.
Security
Prompt injection, least-privilege tool access, sandboxing, audit logs and rate limits. An agent never gets more access than the person it acts for.
Cost and reliability
Token budgets, model selection, caching, retries and fallbacks for agents that run every day.
You’ll leave able to
- Design an agent with a clear, limited scope
- Connect it safely to tools and data
- Test and trace it before release
- Put approvals and logging around real actions
Train your team.
In person or online, for any organisation. Tell us who needs training and which tools they use, and we will tailor the course.
Thank you. Request received.
We will be in touch within two business days.