AI-AGENT DEVELOPMENT

This opening session sets expectations and frames the course from both a technical and career perspective. You’ll understand how the course is structured, what you’ll build, and how agent development skills translate into real-world roles and impact.

• Instructor introduction 
• Audience polls
• Course structure & expectations
• AI job market, agent impact
• Workstation preparation 
• Capstone project brief

This class establishes a shared mental model for what AI agents actually are — and what they are not. You’ll explore how agentic systems differ from traditional automation, where they deliver real value, and how to spot high-impact use cases in modern organisations.

• Evolution of AI agents
• What makes an AI system “agentic” 
• AI Agents vs traditional automation 
• Enterprise tech stack placement
• What agents can and can’t do 
• Characteristics of high-value agent opportunities 
• Case Study: Real world AI Agents success story
• Demo: Building a simple tool-using agent

Self Study Activity (Not graded): Identify three potential agent opportunities in an industry of your choice.

This session breaks down the internal anatomy of a modern AI agent. You’ll learn how perception, reasoning, memory, and action layers work together, and how popular frameworks implement these concepts in production-ready systems.

• Core components of an agent
• Dominant agentic frameworks
• Tooling & integrations
• Interaction flow design
• Trade-offs
• Workshop: Modify a basic AI Agent 

Assignment #1: Build your first AI agent using a provided template.

This lesson focuses on shaping agent behaviour through better prompting and model choices. You’ll learn how reasoning patterns, structure, and LLM selection directly affect reliability, cost, and performance in agentic systems.

• How LLMs process prompts
• Advanced prompt engineering patterns
• Structured outputs & validation
• Reasoning approaches
• Guardrails & error handling
• LLM selection
• Demo: Demonstrate how reasoning changes with varying inputs of prompts of different quality 

Self Study Activity (Not graded): Design a multi-step prompt pipeline and compare outputs.

Here you’ll move beyond stateless agents and enable multi-step autonomy. The focus is on memory systems, retrieval, and tool integration — the foundations required for agents that reason, act, and adapt over time.

• Role of memory 
• Memory design patterns
• Retrieval & RAG
• Sensitive vs public data
• Tool & function calling
• Multi-step workflows
• Workshop: Building a RAG enabled agent

Assignment #2: Integrate tools and memory to the agent from Assignment #1 and include multi step behaviour. 

This lesson introduces systems where multiple agents collaborate to solve complex problems. You’ll learn how to design roles, coordinate communication, and manage the added complexity of multi-agent workflows.

• Task decomposition
• Agent roles & archetypes
• Communication protocols
• Collaboration strategies
• Managing multi-agent complexity
• Demo: Agents collaborating

Assignment #3a:Create a multi-agent system where two/three agents collaborate to produce an output.

This session explores agents that work across text, images, documents, and audio. You’ll evaluate when multimodal capabilities add genuine value and learn how to integrate them responsibly into agent systems.

• What makes an agent multimodal
• Core multimodal tools & libraries
• Designing for multimodal inputs
• Practical constraints
• Industry use cases
• Case Study: Canary Speech: Analysis of auditory biomarkers for medical diagnosis.
• Workshop: Building an agent with basic computer vision capabilities

Assignment #3b: Create a Simple Multimodal Agent and integrate it into the multi-agent system.

This lesson focuses on making agents observable, debuggable, and enterprise-ready. You’ll learn how to log decisions, trace actions, and build systems that can explain why an agent behaved a certain way.

• Designing audit-ready agents
• Structured logging 
• Decision tracing & replay
• Monitoring performance
• Regulatory & enterprise considerations
• Demo: Tracing agent decisions in real time

Understand the processes of verification, validation, and testing in AI agent development to ensure agents function as intended, meet requirements, and perform reliably in real-world scenarios. 

• Best practices for evaluating performance
• Detecting issues
• Refining agents for production readiness

Assignment #4: Add monitoring and testing to your agent, including logs, metrics, and a test script.

Discover how to measure agent quality, identify failure modes, and optimise behaviour over time. You’ll apply practical techniques to make agents more stable, accurate, and cost-effective.

• Performance metrics
• Reducing hallucinations
• Error-handling & fallback strategies 
• A/B testing 
• Latency profiling & token-cost optimisation
• Human-in-the-loop evaluation 
• Demo: A/B testing prompts 

This session covers the security and governance foundations required for deploying agents in sensitive environments. You’ll learn how to protect data, manage access, and reduce attack surfaces in agent workflows.

• Protecting sensitive inputs & outputs 
• Prompt injection & data leakage
• Access control & API management
• Governance structures
• Security monitoring
• Compliance considerations
• Enterprise considerations

Assignment #5: Threat-model your agent: list possible attack vectors and propose mitigations.

This lesson builds awareness and practical skills for identifying and mitigating ethical risks in agentic systems. You’ll apply UK and EU regulatory principles to real-world agent behaviour.

• Bias, fairness & responsible agent behaviours
• Privacy considerations
• Regulatory frameworks
• Detecting & mitigating bias
• Case Study: Analysing an ethical failure scenario in an agentic workflow

You’ll learn how to move from prototype to deployment by packaging, exposing, and operating agents in real environments. The focus is on scalable, maintainable delivery patterns.

• Packaging agents into deployable services 
• Orchestration basics
• Scaling inference workloads 
• Monitoring live agents
• Workshop: Deploy your agent locally using a simple service wrapper

This lesson shifts focus from building agents to selling and adopting them. You’ll learn how to evaluate commercial viability, align with enterprise needs, and articulate value clearly.

• Identifying high-value use cases 
• Data readiness & ROI
• Enterprise adoption patterns
• Workflow integration
• Workshop: Pitch your agent as a product