Skip to main content

Exploratory: Bridge Content Stretch Projects

These are exploratory stretch projects — not required to complete Module 3. They bridge the platform AI evaluation from Module 2 to the agent building work starting in Module 7.


Project 1: Platform AI vs. Custom Agent Decision Matrix

Estimated time: 25 minutes Extends: Module 3 reading (what custom agents add) Prerequisites: Modules 1-3 completed

What You Will Build

A decision matrix for your top 5 operational tasks — for each task, evaluate whether it is better served by platform AI (use what AWS already provides) or a custom agent (build with Hermes). This is the foundation for your Module 4 impact assessment.

Challenge

The decision is not binary, and it changes over time. A task that platform AI handles today may need a custom agent as your requirements grow. The challenge is being honest about what platform AI cannot do — versus what it just does differently than you would prefer.

Steps

  1. List your top 5 operational tasks that involve monitoring, diagnosis, or response (examples: EC2 health monitoring, RDS performance investigation, cost anomaly triage, K8s pod failure response, deployment validation)

  2. For each task, complete a decision row:

TaskPlatform AI Available?What It Can DoWhat It Cannot DoDecisionRationale
EC2 health monitoringYes (CloudWatch)Anomaly detection, alarmsCross-reference runbooks, structured diagnosisCustom agentNeeds context my environment
...
  1. Decision criteria:
    • Platform AI: Task is well-scoped, platform AI handles it accurately, no cross-domain reasoning needed
    • Custom Agent: Task requires domain context (your runbooks, your thresholds), cross-domain reasoning, or structured output in a format the platform doesn't provide
    • Both: Use platform AI as the trigger, custom agent for the analysis (e.g., CloudWatch alarm fires, custom agent investigates)

Expected Deliverable

Completed decision matrix with rationale for each decision, plus a summary: how many of your top 5 tasks benefit from custom agents?


Project 2: First Agent Design Sketch

Estimated time: 20 minutes Extends: Module 3 demo (Hermes first-run walkthrough) Prerequisites: Module 3 completed, familiarity with one of the three Module 10 agent tracks

What You Will Build

A one-page design document for a custom agent that addresses a gap you identified in your Module 2 coverage audit or your Module 3 decision matrix. This is a design sketch — no code, no configuration — just the conceptual design for what the agent would do.

Challenge

Keeping scope narrow. It is tempting to design an agent that does everything. A well-scoped first agent does one thing very well and has clear boundaries for what it does not do. The challenge is resisting scope creep in the design phase.

Steps

Write a one-page design doc using this structure:

# [Agent Name]

## Problem
[One paragraph: what is the manual operational task this agent addresses?
What is the current friction?]

## Agent Scope
- **Does:** [List 3-5 specific actions the agent takes]
- **Does not do:** [List 2-3 explicit out-of-scope items]

## Inputs
[What data does the agent need to do its job?]

## Outputs
[What does the agent produce? Format? Destination?]

## Success Criteria
[How do you know the agent is working well? What would make it worth deploying?]

## Open Questions
[What do you need to figure out before building this?]

Expected Deliverable

A completed one-page design doc. Bring it to Module 4 (Impact Assessment) — it becomes your capstone candidate.


Which Project Should You Do?

Your InterestRecommended Project
Systematic evaluation frameworkProject 1 (decision matrix)
Starting your capstone design earlyProject 2 (agent design sketch)
Under 20 minutesProject 2 — direct preparation for the capstone

Both projects directly feed into Module 4's impact assessment. Project 2 is especially valuable if you want to arrive at Module 4 with a strong capstone candidate already sketched.