Agentic AI Foundations / Module 8
โ† Course Hub
MODULE 8 ยท WRAP-UP

Wrap-Up, Glossary & Cheat Sheet

Course recap, six mental models to keep, cheat tables, and the certification path. Print it โ€” it's your take-home reference.

๐ŸŽ“ You Made It Through the Day

You can nowโ€ฆ

๐Ÿง  Six Mental Models to Keep

๐Ÿ”„

Perceive โ†’ Reason โ†’ Act

Every agent is this loop. It reads inputs, decides over multiple steps, then does something โ€” and repeats. A chatbot stops at "reason"; an agent closes the loop with action.

๐ŸŽš๏ธ

Match autonomy to risk

Assisted โ†’ Workflow-automation โ†’ Partial โ†’ Full. The higher the stakes (client money, compliance, regulatory reporting), the lower the autonomy and the firmer the human gate. Finance earns its trust slowly.

๐Ÿงฉ

Four patterns cover most needs

Chaining (do it in order) ยท Routing (send it the right way) ยท Parallelisation (do many at once) ยท Orchestration (a coordinator spawns sub-agents). Name the pattern before you build.

๐Ÿ—๏ธ

The Quick stack is 4 layers

Space (the kitchen) โ†’ Chat Agent (the recipe) โ†’ Flow/Schedule (the doorbell) โ†’ MCP (the oven). 80% of value lives in the first two layers.

๐Ÿ›ก๏ธ

Agents earn trust through guardrails

Named "never do" rules, human-in-the-loop triggers, traceable numbers, and an audit trail. Flag-only beats auto-decide for anything an auditor or regulator will inspect.

๐Ÿ”

Misses are spec gaps, not AI failures

When an agent misses something, don't fix it in-chat โ€” add the rule to the agent definition. The agent gets smarter for everyone on the next run.

๐Ÿ“– Glossary

TermMeaning
SpaceContainer holding documents and scoping an agent's knowledge โ€” "the kitchen"
Chat AgentA saved, reusable prompt configuration with persona, instructions, knowledge, actions, and suggested prompts โ€” "the recipe"
FlowA no-code, multi-step wrapper around one or more Chat Agents, schedulable and shareable โ€” "the doorbell"
MCP ConnectorModel Context Protocol integration giving an agent live tool access to external systems โ€” "the oven"
Quick ResearchAn AI-led, multi-step investigation feature that produces a cited report
Pilot CanvasThe capstone deliverable โ€” a structured one-pager describing a proposed real agentic pilot, scored on autonomy/tools/guardrails/impact
Rule of ThreeHeuristic: typed the same prompt 3 times โ†’ save it as an agent
Ghost AgentA failure mode โ€” an agent built once, shared, then silently abandoned/never updated
Audit-DefensibilityThe finance-specific design pattern: named output sections, traceable numbers, flag-only behavior, cited sources
Break (reconciliation)A line item where two records (e.g., GL and bank statement) don't match
NCR-style triageBorrowed term from manufacturing quality processes โ€” classifying and routing an exception for review, never auto-resolving it

๐Ÿงญ Which Quick Surface Do I Use?

NeedUse
Answer a policy question with citations, repeatedlyChat Agent
Run the same multi-step process on a scheduleFlow
One-off deep investigation with cited sourcesQuick Research
Connect to a real system (core banking, market data)MCP Connector

๐Ÿš€ Keep Learning

๐Ÿ“š

AWS Skill Builder

600+ free self-paced digital courses, labs, and learning plans.

๐ŸŽ“

AWS Certified AI Practitioner

The foundational credential validating GenAI & AI fundamentals โ€” the natural next cert after this course.

๐Ÿ…

Certification Path

Foundational โ†’ Associate โ†’ Professional โ†’ Specialty. Start Foundational; no prior experience needed.

๐Ÿค

Your AWS Account Team

Ready to turn your Pilot Canvas into a real pilot? Your account team can help scope data, security, and a production path.