01 · Prototype
Leadership Simulator
01
React.js
Anthropic Claude API
Agentic NPC Design
Real-time State
A four-round leadership simulation demo where the user inherits a team damaged by a toxic predecessor. Four NPC team members, each with distinct archetypes and live morale, trust, and engagement stats, respond dynamically to every decision via AI-generated reactions. Bidirectional leadership style meters (Directive vs Empowering, Hardline vs Flexible, Abrasive vs Listening, Results vs People) build a real-time profile that culminates in a leadership archetype diagnosis.
Built before there was a business case for it, to test whether agentic AI could deliver meaningfully better experiential leadership development than scripted branching scenarios. The honest answer is yes, but only when the prompt architecture, fallback design, and state model are treated with the same care we would give to any production system.
Video walkthrough
Loom embed loads here · approx. 3 min
What it demonstrates
End-to-end build of a working agentic application: API integration, prompt design across multiple character voices, state management, an eight-second timeout with graceful fallback handling, and consequential-decision learning mechanics inspired by the mechanical depth of Football Manager.
02 · Prototype
Multi-Agent Instructional Design Pipeline
02
AutoGen Studio
Multi-Agent Orchestration
Needs Analysis
Workflow Design
A three-agent system built in AutoGen Studio that takes a raw learning need from intake through to an instructional design brief. Three specialised agents, armed with different LLMs, handle needs analysis, design specification, and brief generation, with structured handoffs between roles.
Not production-grade by design. The intent was to test where agentic patterns hold up and where they break down when applied to L&D workflows that are normally handled through sequential human steps. The handoff scoping question turned out to be more interesting than the agents themselves.
What it demonstrates
Agent role design, handoff scoping, and multi-agent orchestration applied to a domain (L&D process design) that is usually handled by humans doing sequential cognitive work. Also: comfort moving between vendor stacks, not just Anthropic.
03 · Writing
Ground 0 / 1 / 2 · Agentic AI Maturity in L&D
03
Substack · Learning Adept
Operational Framing
Maturity Model
A long-form piece on where agentic AI shows up in learning and development, and where the gap between AI adoption and real operational change actually lives. The piece introduces a Ground 0 / Ground 1 / Ground 2 maturity model: AI as content generator, AI as embedded assistant, and AI as autonomous agent inside the learning architecture itself.
The argument: most organisations are stuck at Ground 0 and calling it transformation. Real change shows up when AI moves from production tool to operational infrastructure, which is a redesign problem, not a tooling problem.
Read on Learning Adept
→
What it demonstrates
How I think about AI as operational redesign rather than tool deployment, and the frameworks I use to diagnose where organisations actually are versus where they think they are.