AI for Everyone: A Practical Course for Non-Technical People

By Agentive (www.agentive.au)

Audience: business owners, managers and staff who use AI (or want to) but are not technical. Goal: attendees understand how AI really works, use it safely and confidently, and know what tools, APIs, MCP and skills mean for their business. Format: 4 sessions of 90 minutes (or 2 half-days of 3 hours). Each session works on its own as a standalone workshop. Style: plain language, everyday analogies, live demos, hands-on exercises. No coding required.

Course at a glance

Session Title Core topics
1 How AI Thinks LLMs, training, tokens, context window, models, randomness
2 Talking to AI Well Prompting, context, hallucination, checking answers
3 AI That Does Things Harness, tool calling, APIs, MCP, agents, skills
4 Using AI Safely at Work Golden rules, privacy, security, prompt injection, approvals, your AI plan

Short options - 90-minute taster: Session 1 (condensed to 30 min) + Session 2 hallucination segment + Session 3 overview + the Golden Rules. - 2-hour workshop: the taster plus one hands-on exercise and Q&A.

Session 1: How AI Thinks (90 min)

Outcome: attendees can explain in one sentence what an AI chatbot actually does, and why it has limits.

Time Topic Key messages Activity
0:00 to 0:10 Welcome and myth-busting AI is not a person, a search engine or a database. It is a pattern-prediction machine. Poll: "What confuses or worries you about AI?"
0:10 to 0:25 What is an LLM? A Large Language Model learned from a huge amount of text and predicts the most likely next word. Analogy: autocomplete with a library in its head. Live demo: "Finish this sentence..." with different starters
0:25 to 0:35 Training and knowledge cutoff The model learned up to a certain date. It does not know today's news unless given a tool to search. Ask AI about a very recent event, with and without web search
0:35 to 0:50 Tokens AI reads text in small chunks called tokens (roughly three quarters of a word). Tokens drive cost, speed and limits. Show a sentence split into tokens; guess the count
0:50 to 1:05 Context window The AI's short-term memory: everything it can "see" right now. When it fills up, old details drop off or get summarised. Analogy: a desk that only holds so many pages. Long chat demo: AI forgets an early instruction
1:05 to 1:15 Models and randomness Different models have different strengths, speeds and costs. Same question can give different answers (temperature). Ask the same question three times and compare
1:15 to 1:30 Recap and Q&A The five words: LLM, token, context, model, cutoff. Quick quiz (Kahoot style)

Session 2: Talking to AI Well (90 min)

Outcome: attendees write clear prompts, give useful context and catch wrong answers.

Time Topic Key messages Activity
0:00 to 0:15 Prompting basics A good prompt has Role, Task, Context, Format (RTCF). Treat AI like a smart new staff member on day one. Rewrite a weak prompt into a strong one
0:15 to 0:30 Giving context The AI only knows what you tell it. Paste the email, explain the audience, give examples of "good". Same task with and without context; compare
0:30 to 0:40 Iterating The first answer is a draft. Ask it to shorten, change tone, try again, explain its reasoning. Three rounds of refinement on one email
0:40 to 1:00 Hallucination AI can be confident and wrong. It invents facts, quotes, figures, legal references and links because it predicts plausible text, not truth. "Spot the lie": find made-up facts in AI answers
1:00 to 1:15 Checking answers Ask for sources, verify numbers, cross-check important facts, never trust a link you have not opened. Verification checklist practice
1:15 to 1:30 Hands-on Use AI on a real task from your own work. Pairs share before and after results

Prompt template handout

You are a [role]. I need you to [task]. Here is the background: [context]. The audience is [who]. Please give me [format, length, tone]. If anything is unclear, ask me before answering.

Session 3: AI That Does Things (90 min)

Outcome: attendees understand the difference between a chatbot that talks and an AI agent that acts, and what tools, APIs, MCP and skills are.

Time Topic Key messages Activity
0:00 to 0:15 The harness The model is the engine; the harness is the car around it: instructions, memory, tools, files, rules and permissions. ChatGPT, Claude and AI employees are all model + harness. Diagram: engine vs car
0:15 to 0:30 Tool calling AI cannot click buttons by itself. It asks the harness to run a tool ("search the web", "read this file", "send email"), gets the result, then continues. Analogy: a manager asking an assistant to fetch a file. Live demo: watch an AI call a calculator or web search step by step
0:30 to 0:45 APIs An API is a doorway that lets one piece of software talk to another (Xero, Gmail, HubSpot). Analogy: a restaurant waiter taking orders to the kitchen. API keys are like passwords: keep them secret. Map: which of your business apps have APIs?
0:45 to 1:00 MCP (Model Context Protocol) A universal plug standard that lets AI connect to many apps the same way. Analogy: USB-C for AI. Build the connector once, any compatible AI can use it. Show an AI connected to a calendar or CRM via MCP
1:00 to 1:15 Skills and agents A skill is a packaged "how-to" that teaches AI a specific job (e.g. "create an invoice the way we do it"). An agent combines model + tools + skills + memory to complete multi-step work. Walk through an AI employee doing a real task end to end
1:15 to 1:30 Recap and Q&A Chatbot = talks. Agent = acts. More power means more need for rules (Session 4). Match-the-term game

Simple picture to show

You  →  Harness (rules, memory, skills)  →  Model (the brain)
                    ↓
          Tools via APIs / MCP  →  Gmail, Xero, HubSpot, files, web

Session 4: Using AI Safely at Work (90 min)

Outcome: attendees leave with clear rules, know the main risks and have a personal plan for using AI.

Time Topic Key messages Activity
0:00 to 0:15 Privacy and data What you type may be stored or used for training, depending on the product and plan. Business plans and private AI employees differ from free tools. Sort examples: safe to paste or not?
0:15 to 0:30 Security risks Prompt injection: hidden instructions inside emails, documents or web pages can trick an AI. Leaked API keys. Fake AI tools and plugins. Demo: a document with a hidden instruction
0:30 to 0:45 Human in the loop AI drafts, humans approve. Anything that sends, pays, publishes or deletes needs a person to sign off. Decide: which tasks need approval?
0:45 to 1:00 Bias, copyright and disclosure AI can reflect bias, reuse others' work and should be disclosed where it matters (clients, regulators). Discussion of real scenarios
1:00 to 1:15 The Golden Rules See below. Group agrees on a team AI policy
1:15 to 1:30 Your AI action plan Pick three tasks to start with this week. Personal plan worksheet

The Golden Rules for Using AI

Protect 1. Never paste passwords, API keys, tax file numbers, bank details or health records into a public AI tool. 2. Use business-grade or private AI for client and company data. 3. Only connect apps and install skills or plugins from trusted sources. 4. Treat emails, attachments and web pages as possible carriers of hidden instructions.

Verify

  1. Assume every fact, number, quote, link and legal reference could be wrong until checked.
  2. Ask the AI for its sources, then open them yourself.
  3. Double-check anything financial, legal, medical or compliance related with a qualified person.

Control

  1. AI drafts, you decide. A human approves anything that is sent, paid, published or deleted.
  2. Give AI the least access it needs (read-only before write access).
  3. Keep a record of what AI did on important tasks.

Use well

  1. Give context: the AI only knows what you tell it.
  2. Be specific about the task, audience, format and tone.
  3. Start a fresh chat for a new topic so old context does not confuse it.
  4. Iterate: the first answer is a draft, not the final product.
  5. Be open with clients and colleagues when AI has been used.

Glossary (handout)

Term Plain English
AI Software that performs tasks that normally need human thinking.
LLM Large Language Model: the "brain" that predicts text.
Token A small chunk of text the AI reads and writes; the unit of cost and limits.
Context window How much the AI can hold in mind at once.
Knowledge cutoff The date the model's training data stops.
Prompt The instruction you give the AI.
Hallucination A confident but made-up answer.
Harness The software around the model: rules, memory, tools, permissions.
Tool calling The AI asking its harness to perform an action and return the result.
API A doorway that lets software talk to other software.
API key A secret password for an API. Never share it.
MCP Model Context Protocol: a standard plug to connect AI to apps and data.
Skill A packaged set of instructions that teaches AI a specific job.
Agent AI that plans and takes multi-step actions using tools.
Prompt injection Hidden instructions in content that try to hijack an AI.
Human in the loop A person reviews and approves AI work before it takes effect.

Materials

Facilitator notes