Workshop 3 · Session agenda

Four hours, building agents and learning what they break.

From "what is an agent" through five hands-on builds and a documented failure mode you take home.

Before you start

Open this site on your laptop.
Bookmark /live in a second tab.
Have free-tier accounts ready for: Claude or ChatGPT, Zapier or Make, and a Google account (for the spreadsheet sandbox).
Workshop 3 assumes you know RTCF prompts (Workshop 1) and tech triage (Workshop 2). If not, it will still work, but plan to study those after.

The agenda

0:00
Welcome and where this fits
Workshop 3 of three. Why "agent" is the buzzword and what it actually means. Honest framing: most agentic demos work 70% of the time, and 70% is unsafe for any task with cost.
0:15
The agent landscape, 2026 edition
Custom GPTs, Claude Projects, Poe bots, Zapier and Make, browser agents, Manus. Free vs. free-with-account vs. paid-only. What each is good at, what each fails at.
0:35
Live demo: build a custom bot in 5 minutes
From empty Claude Project to working tutor bot, on the projector. Watch what changes when the system prompt is sloppy versus tight.
0:50
Activity 1: build your CompTIA tutor bot
30 minutes. You leave with a bot you can use tonight. /ws3-agentic/activities
1:25
Break
10 minutes.
1:35
Live demo: the chained-prompt automation
Build a Google-Sheet-to-AI-to-cover-letter Zap on the projector. Show the failure when the prompt is loose; show the fix.
1:50
Activity 2: job-search agent
35 minutes. Wire Zapier or Make to pull postings, extract ATS keywords, draft a letter. Human-in-the-loop is the whole point.
2:30
Activity 3: help desk auto-triage flow
35 minutes. Same automation pattern, applied to triage. Build against a sandbox; do NOT wire to your real ticketing system.
3:05
Break
10 minutes.
3:15
Activity 4: browser agent on a real research task
30 minutes. Run Perplexity on an employer-research task. Verify three claims. Write the safety review.
3:45
Activity 5: evaluate one of your agents
25 minutes. Pick any agent you built today. Run four tests. Document one failure mode card.
3:55
Wrap and where to next
The 30-day plan. The cost-and-limits use case. Open mic for questions about your specific agent ideas.

The agenda fits five activities into four hours by design — not all participants finish all five. Activities 1, 2, and 5 are the must-do trio; 3 and 4 can be take-home.

What to bring

A laptop. The agentic builds do not work well on a phone.
Free-tier accounts on at least Claude or ChatGPT, plus Zapier or Make, plus Google (for the sandbox sheet).
Optional: a real CompTIA exam date you are studying for, a real city you are job-searching in, a real complaint pattern from your help desk work. The activities are richer when grounded in your real life.