Course 2 · Intermediate · 2 days

AI at Scale.

Two days for advanced AI users who want to move from chat to agents, workflows and team-scale automations. Build a reusable agent, a workflow with a human approval gate, a two-agent system with shared memory — and stress-test it against the adversarial inputs that break production systems.

Cape Town, in-person only · two consecutive days · Max 14 seats · 6 hands-on exercises + capstone · Certificate ·

Who it's for

You learned to use AI. Now learn to build with it.

If "AI at Work" is about getting comfortable with prompts, this course is about moving from chat to agents and workflows — production-grade systems with roles, tools, memory, human approval gates, and the three checks (factuality, safety, cost) that decide whether they actually ship.

The advanced individual contributor

You're already prompting daily. You want to stop re-typing prompts and start shipping reusable agents — with the discipline of role, tools, memory and loop.

The operations / ops-tech owner

You spot the same repetitive workflow five times a week. You want to wire it to a trigger, hand it to one or two agents, and put a human gate where reversibility ends.

The team lead who'll roll it out

You'll leave with a Monday-morning team workflow — three agents, one shared memory, one digest, one human gate — that survives adversarial inputs.

Prerequisite: AI at Work (or equivalent) and one licensed AI product — Claude Pro / Copilot for M365 / Gemini Advanced / Notion AI / NotebookLM. The trainer demoes every exercise in Claude live; students follow their own tool's path.

What you'll leave with

Six working artefacts. One Monday-morning team workflow. One running case.

Running case: Karoo Logistics — Sam Naidoo (now Operations Director) has been asked for three AI automations the team uses every Monday. You build them.

  • A reusable agent — Role · Tools · Memory · Loop. Sam's weekly KPI digest, same agent each Monday, fresh inputs every week.
  • A workflow with a human approval gate — complaint triage that classifies, drafts, holds, then sends only when a person says go.
  • A live link from agent to real data — 40 supplier emails a week, surface the 5 that need a reply. Native / no-code / MCP integration patterns covered.
  • A two-agent system that hands off — a Researcher reads context + policy + history, a Writer drafts in three tones. A clean hand-off schema.
  • Shared memory across all three agents — one source of truth (Notion DB / Google Sheet / Claude project knowledge) all three read and write.
  • A stress-tested, production-aware version — survives adversarial inputs (missing data, non-English emails, prompt injection). The 3 production checks (Factuality · Safety · Cost) applied.
  • A capstone workflow — all three agents + shared memory wired into one Monday 08:00 trigger, three parallel runs, one digest, one human gate.
  • A laminated reference card — agent anatomy, workflow anatomy, orchestration patterns, the production-check list.
  • A Proven Path certificate of completion — PPA-AIS-001, issued after a 10-question knowledge check (7/10 to pass).
Two days, hour by hour

Agenda — Day 1: prompts to working agents · Day 2: team-scale, memory & safety

Day 1 takes you from one-off prompts to working agents and workflows. Day 2 ties them together with memory, safety and a Monday-morning team workflow. Lunch is BYO during the 45-minute break on each day — V&A Waterfront cafés 5 minutes away.

TimeBlock
Day 1 — From prompts to working agents (09:00–15:00)
09:00 – 09:15Welcome · check-in · Karoo running case
09:15 – 09:35Theory 1. The agent spectrum — prompt → chained prompt → agent → agent team
09:35 – 09:55Theory 2. Anatomy of an agent — Role · Tools · Memory · Loop
09:55 – 10:50Exercise 1. Build Karoo's weekly KPI digest — your first reusable agent
10:50 – 11:05Coffee
11:05 – 11:25Theory 3. Workflow anatomy — Trigger · Input · AI step · Action · Human gate
11:25 – 12:20Exercise 2. Complaint triage with a human approval gate
12:20 – 13:05BYO lunch (45 min)
13:05 – 13:25Theory 4. Tool integration patterns — native · no-code · MCP
13:25 – 14:30Exercise 3. Supplier-email digest — connect your agent to real inbox data
14:30 – 15:00Day-1 reflection · homework
Day 2 — Team-scale: orchestration, memory, safety (09:00–15:00)
09:00 – 09:15Day-2 kickoff · show & tell from Day 1
09:15 – 09:35Theory 5. Multi-agent orchestration — sequential · parallel · supervisor / worker
09:35 – 10:30Exercise 4. Researcher → Writer — a 2-agent hand-off system
10:30 – 10:45Coffee
10:45 – 11:05Theory 6. Memory and state — short-term · long-term · shared
11:05 – 12:00Exercise 5. Shared memory across the three Karoo agents
12:00 – 12:45BYO lunch (45 min)
12:45 – 13:05Theory 7. The 3 production checks — Factuality · Safety · Cost
13:05 – 14:00Exercise 6. Stress-test and harden against adversarial inputs
14:00 – 14:15Theory 8. Cost, latency and governance — when it leaves your laptop
14:15 – 14:40Capstone. Wire everything into one Monday-morning team workflow
14:40 – 14:55Final test — 10 questions, 7/10 to pass
14:55 – 15:00Reflection · certificates · close
The toolkit

Bring one licensed tool. The trainer is on Claude — same exercise, your path.

Every exercise has a tool-agnostic scenario + five tool-specific step-by-steps. The trainer demoes every step in Claude live; when a student's path stalls we triage against the Claude reference.

Anthropic Claude (Pro / Teams)

Projects + Cowork mode (file system access). Long-context drafting, agent loops, code. The trainer's fallback for every exercise.

Microsoft Copilot for M365

Copilot Studio for agents inside the tenant; native Outlook / Word / Excel / Teams actions. Best when work stays inside M365.

Google Gemini Advanced / Gems

Custom Gems + Gmail / Docs side-panels + light AppSheet automation. Best when work lives in Google Workspace.

Notion AI / Business

AI Buttons, AI auto-fill DB properties, DB relations as the agent surface. Your team workspace becomes the agent surface.

Google NotebookLM

Source-grounded notebooks; "Briefing doc" custom output as the workflow output. Best when answers must cite your own documents.

Open-source & MCP intro

Theory beat 4 covers Model Context Protocol — the new standard for agents calling your tools — and where no-code (Make / Zapier / n8n / Power Automate) fits.

Investment

R5,000 TTC in-person per seat.

≈ €250 / $250 per seat — indicative, divided at R20 = €1 / $1.

Two full days (09:00–15:00 each), materials, certificate. Lunch is BYO during the 45-minute break — coffee, tea and a mid-morning snack are. In-person only — the two days work because the room works.

What's included

  • Two full days of training (09:00–15:00 each)
  • Coffee, tea, water and a mid-morning snack each day (lunch not provided — BYO)
  • Student exercise pack + Karoo Data Pack + adversarial inputs
  • Laminated reference card (agent + workflow anatomy, orchestration patterns)
  • 10-MCQ knowledge check + PPA-AIS-001 certificate
  • Email follow-up access for two weeks after the course

Group & in-house

Group of 4+: 10% off the standard rate. In-house: contact us for a quote — same curriculum, your office, your team.

Email us for a quote

International participants & non-Rand payments welcome — we'll send specific invoicing instructions on request.