Agentic AI
Exercises
- Day 1 Exercise — Deciding What AI Should Not Do Place FitPath's proposed features on the AI Capability Ladder, decline at least three on architectural grounds, and pressure-test your reasoning against an LLM arguing the other side.
- Day 2 Exercises — Prompt Engineering Turn Day 1's workout-plan sketch into a structured prompt that actually behaves, measure it, and then try to break it.
- Day 3 Exercises — Knowledge Systems, Vector Databases, and RAG Ground the Day 2 prompt in FitPath's real documents with RAG, then check whether that grounding made it trustworthy or just made it look that way.
- Day 3 Extra Content — Building the RAG Pipeline in Flowise Assemble the RAG pipeline yourself node by node in Flowise — chunking, embeddings, vector store, retriever — instead of letting NotebookLM decide it for you.
- Day 4 Exercises — Agentic AI Foundations and Agent Design Give FitPath's Weekly Check-In feature real tool access, watch an ungoverned agent make a judgment call on a genuinely risky case, then add an approval gate and confirm it actually changes what the agent does.
- Day 5 Exercise — Build a Supervisor Agent with Specialized Sub-Agents Split a single check-in agent into a supervisor and three specialized sub-agents, wire delegation and a human checkpoint in Fleet, then prove it behaves differently for a clean user than a flagged one.
- Day 6 Exercises — Enterprise Agent Adoption and Business Value Score FitPath's AI portfolio on value and feasibility, build a full TCO and value case for your Day 4 agent, then run discovery on a stakeholder who won't give you a straight answer.
- Day 7 — Exercise 1: The Nutrition Agent (A Real A2A Handshake) Build a standalone Nutrition Agent with its own Agent Card, then a Workout Agent stand-in that discovers it and hands it a real task over A2A — the peer-agent alternative to bolting a nutrition tool onto your MCP server.
- Day 8 — Exercises: Poisoned Tool Output & Build the Audit Record Watch your own agent get manipulated by poisoned tool output, add one real defense against the injection, then build the structured 8-field audit record that proves what happened and who signed off on it.
- Day 9 — Exercise: The Full Autonomy Ask Run a full customer conversation with a VP who wants to fully automate member support in two quarters — discovery, objection handling, and a written recommendation that pushes back on the parts of the ask that don't hold up.
- Day 10 — Priya's Next Ask: The Trainer Assistant Draft an architecture for an AI Trainer Assistant, then defend it live against a VP who pushes back on every human-approval step — and revise your write-action boundary after she asks the question you should have already answered.
- Day 11 Exercise: The Category Switch Hold one live conversation with a VP who opens with a coding-agent challenge (Mentor vs. Cursor) then pivots without warning to a framework question (Agent Workbench vs. LangGraph) — reusing your first answer is the main way to fail.
- Day 12 Exercises — Strategic AI Adoption with OutSystems Trace one FitPath capability's architecture and governance evolution across the course, then run a cold customer conversation with no persona history to lean on — closing with a written adoption roadmap memo.