Programme Content
Module 1: Understanding agentic AI (taxonomy, memory, RAG)
Structured debrief of pre-work agent interactions. Surfacing assumptions and establishing shared vocabulary.
Hands-on laboratory: rotate through four agent types — simple assistant, task agent with tools, autonomous research agent, and a deliberately ‘agentwashed’ tool. Observe reasoning, tool use, and memory in action.
Agent taxonomy: assistants, task agents, autonomous agents, and multi-agent systems. Decision framework for selecting the right approach.
Agent memory architectures and their strategic implications. Introduction to data readiness and RAG as the bridge between agents and enterprise knowledge.
Module 2: Navigating hype (diagnostic framework for ‘agentwashing’)
The ‘agentwashing’ phenomenon: a diagnostic framework for evaluating vendor claims. This lens applies throughout the remainder of the programme.
Live failure demonstrations: agents hallucinating, failing silently, going off-rails, amplifying bias, and leaking data. Distinguishing fundamental limits from areas of rapid improvement.
Module 3: Building and connecting (hands-on lab: MCP/A2A protocols)
Hands-on laboratory: build an agent for a business task from your pre-work brief. Configure skills and tools using MCP-style connections. Connect to data sources via RAG. Iterate as design choices shape behaviour.
Protocols and standards: MCP (Anthropic), A2A (Google), and ACP (IBM). Strategic implications for interoperability, vendor lock-in, and platform architecture.
Enterprise case study and practitioner panel: what worked, what failed, the real costs, and the workflow redesign required.
Module 4: Operational governance (lab: stress-testing and breaking agents)
Hands-on laboratory: systematically break your Day 1 agent. Adversarial inputs, prompt injection, edge cases, and tool manipulation. Design governance protocols in response.
Governance and trust frameworks: bounded autonomy, audit trails, and explainability. A practical four-level autonomy framework from human-directed to fully autonomous. Human-in-the-loop versus human-on-the-loop.
EU AI Act and emerging UK regulatory landscape. Positioning governance as an enabler of deployment.
Module 5: From pilot to production (observability, ROI, AgentOps)
Observability and agent operations: tracing multi-step reasoning, evaluation frameworks, cost monitoring, latency tracking, and regression detection. The observability stack: what to build, what to buy.
Economics of agentic AI: real deployment costs, cost optimisation as an architectural discipline, ROI measurement. The emerging agent ecosystem.
Workflow redesign versus agent overlay: why overlaying agents on legacy processes rarely succeeds at scale.
Module 6: Leading the agentic organisation (90-day action plan)
The tool-coworker challenge: agents defy the traditional tool-versus-worker binary. Four tensions leaders must navigate. Designing operating models for human-agent collaboration.
Skills and talent strategy: three layers of capability (agent skills, team skills, leadership skills). New roles and build-versus-hire-versus-upskill decisions.
Facilitated roadmap session: peer groups develop a 90-day action plan. Faculty and peer challenge. Presentations and close.
