Lesson 1
What AI assistants are good at, and where they trip up
What you'll learn: what AI assistants genuinely help with at work, where they reliably go wrong, and how to sort your own to-do list into good and bad fits.
Meet Rosa and her week
Rosa is the office manager at Hollins Landscape Design, a firm of 12 people who design and build gardens. She handles everything that is not digging or drawing: client emails, supplier orders, the newsletter, meeting notes, the monthly report for the two directors, and a steady stream of small surprises.
Her firm has just given everyone access to an AI assistant. Rosa has tried it a few times and found it hit and miss. One day it wrote a lovely email in seconds; the next it produced three paragraphs of waffle and got a date wrong. She has decided to give it a fair trial with one real week of work, and this course follows that week.
On Monday morning, her list looks like this:
- Reply to Mr Okafor, a client who is upset that his patio is two weeks late.
- Write the spring newsletter for past and current clients.
- Turn Tuesday's 90-minute project meeting into notes and actions.
- Compare three quotes from paving suppliers.
- Produce the monthly report for the directors.
What we mean by an AI assistant
By AI assistant we mean a chat tool built on a large language model: ChatGPT, Claude, Gemini, Microsoft Copilot and others. You type a request in ordinary language and it replies in writing. Many can also read files you upload, search the web, or work inside email and document software, but these features vary between products and plans and change often. This course sticks to skills that work in any of them.
One thing is worth knowing about how they work. An assistant writes its reply by predicting, word by word, what a helpful answer would most likely say, based on patterns learned from huge amounts of text. That is why it is so fluent. It is also why it can produce something that sounds right and is not: it is aiming for plausible, and plausible is usually, but not always, correct.
A useful way to think about it
The most helpful picture is a bright temp on their first day. They are quick, well read, never tired and happy to redo a task five times. They can write in almost any style and summarise a long document in moments.
But they know nothing about your firm. They have never met Mr Okafor, have not seen the paving quotes and do not know that the directors hate long reports. If you give them a vague instruction, they fill the gaps with sensible-sounding guesses. And they will rarely say "I'm not sure"; they hand over their work with the same confidence whether it is right or wrong.
Treat the assistant like that temp. Brief it well, check its work, and never let it send anything in your name unread.
Where assistants shine
Assistants are consistently good at:
- Drafting: emails, letters, announcements, job adverts, first versions of almost anything.
- Rewriting: shorter, friendlier, more formal, simpler, in a different structure.
- Summarising: pulling the main points out of long text you provide.
- Brainstorming: options, angles, questions you had not thought of.
- Explaining: a confusing clause, a technical term, a spreadsheet formula.
- Structuring: turning messy notes into a tidy list, table or plan.
Where they trip up
They are much weaker at:
- Facts they were not given: your prices, your dates, your clients. Anything specific to your work must come from you.
- Recent events: their knowledge stops at a training cutoff unless a web search feature is switched on and used.
- Precise arithmetic on lots of numbers: they can slip, especially with long tables.
- Exact quotes and references: they can invent a plausible-looking source.
- Judgement calls about people: they do not know your colleagues or clients.
Sorting Rosa's list
Rosa sorts her week by asking two questions of each task: is this mostly writing or rearranging information I can supply, and how bad would a mistake be?
| Task | Good fit? | Why |
|---|---|---|
| Reply to Mr Okafor | Yes, with care | Drafting is a strength; the facts and tone must be hers |
| Spring newsletter | Yes | Drafting from her notes; she checks every claim |
| Meeting notes and actions | Yes | Summarising a transcript she provides |
| Comparing paving quotes | Partly | Good for laying out differences; totals need checking |
| Monthly report | Partly | Good for structure and wording; figures must come from the accounts |
None of the tasks is a straight "no", but two come with a clear warning. Rosa also notices that nothing on the list is "find out a fact I don't know", which is where assistants are least trustworthy.
What a good week will look like
By the end of this course, Rosa will have:
- A way of writing requests that gets a useful first draft most of the time.
- A habit of improving drafts in a few quick rounds rather than starting over.
- Methods for drafting, summarising and comparing that keep her in control of the facts.
- A checklist for anything that leaves her desk.
- Clear rules for what she never pastes into an assistant.
- A small library of reusable prompts for the jobs she does every month.
Recap
- An AI assistant is a chat tool built on a language model; product names and features vary and change.
- Think of it as a bright temp on their first day: fast and capable, but knowing nothing about your work unless told.
- It is strong at drafting, rewriting, summarising, brainstorming, explaining and structuring.
- It is weak at facts you did not supply, recent events, heavy arithmetic, exact references and judgement about people.
- Sort tasks by whether you can supply the information and how costly a mistake would be.