AI Fundamentals · Lesson 1 · about 20 minutes
What AI Can and Cannot Do for Your Work
AI will not take over your job this year, but it can take over the parts of your job that are slow and repetitive. The skill is knowing which parts.
What a chat AI tool actually is
ChatGPT (from OpenAI), Claude (from Anthropic), Gemini (from Google) and Microsoft Copilot are all built on large language models. A large language model is trained on a very large amount of text and learns the patterns of language. When you type a request, it produces a reply one piece at a time, each time predicting what is most likely to come next.
That simple idea explains both its strengths and its weaknesses:
- It is very good at language work: drafting, rewording, summarising, translating, changing tone, structuring messy notes.
- It does not check facts the way a person with a reference book does. It produces text that sounds right. Most of the time that text is also correct, but not always.
Where AI helps at work
AI earns its keep on tasks that are mostly reading and writing, where you can check the result quickly:
| Task | What AI does well | What you still do |
|---|---|---|
| Emails and letters | First draft in seconds, in the tone you ask for | Check names, figures and promises |
| Long documents | Summary of the key points and actions | Read the parts that matter before acting |
| Meeting notes | Turns rough notes into minutes and a task list | Confirm who agreed to what |
| Reports | Structure, headings, plain-English rewrites | Supply the real numbers |
| Spreadsheets | Explains and writes formulas | Test the formula on known data |
| Customer replies | Polite, consistent answers to common questions | Approve anything about money or complaints |
Where AI gets things wrong
When a model does not know something, it can still produce a confident, detailed answer. People call this a hallucination. Common examples are invented statistics, made-up references, wrong dates, and laws or prices that have changed since the model was trained.
Three habits protect you:
- Never repeat a number or a fact you have not checked against a source you trust, such as KNBS, the Kenya Revenue Authority or the official site of the organisation concerned.
- Give the tool your facts. If you paste in the real figures and ask it to write around them, it has far less room to invent.
- Match the checking to the risk. A social media caption needs a quick read. A client contract, a tax figure or a medical instruction needs a qualified person.
A simple rule for the rest of this course
Use AI to produce the first 80 percent of a piece of work, then spend your own time on the last 20 percent: the facts, the judgement and the relationship. That is where your value is, and it is the part AI cannot be trusted with.
If you remember one thing from this lesson: AI drafts, you decide.
Key takeaways
- AI tools like ChatGPT predict likely words; they do not look facts up, so they can sound sure and still be wrong.
- AI is strongest on reading and writing work you can check quickly: drafts, summaries, minutes and rewrites.
- Never repeat a statistic or legal point from an AI chat until you have checked the original source.
- Let AI produce the first draft and spend your own time on the facts, the judgement and the client.
Your task this week
Write down five tasks you did last week that involved reading, writing or summarising. Next to each, note whether a mistake would be cheap (easy to spot and fix) or costly (money, a client, the law). Keep the list: lesson 8 turns it into your plan.
Check your understanding
Three quick questions. Your answers stay on your device.