What AI is actually good at, in your day
Three jobs it does well, two it really doesn't.
By the end of this lesson
After this lesson, you'll be able to name three jobs AI is genuinely useful for in your role — and two it'll quietly get wrong if you let it.
Most workplaces rolled out AI in 2024 and 2025 with a one-line announcement: "Copilot is now available across the company." Then nothing. No training. No examples. No rules. You were left to figure out what it's for on your own.
Here's the version you didn't get sent. AI is great at a small number of things and quietly bad at others. Knowing the difference is the whole game.
Three jobs it actually does well
For most non-technical office work, three categories of task are where AI clearly earns its keep.
- First drafts. Status reports, customer-facing emails, slide structure for Friday's presentation, a job description, a meeting agenda, a short policy paragraph. Anything where staring at a blank page used to cost you twenty minutes.
- Summarising long things. A 90-minute meeting transcript turned into five bullets. A 200-page regulator submission turned into the three pages your team actually needs to read. A long email thread turned into "here's what's been agreed and here's what's still open."
- Comparing two things. Two supplier proposals side-by-side. Last quarter's report against this quarter's. Your draft policy against the new compliance rules. AI is good at noticing differences a tired human eye glides past.
If most of your week is producing documents, summarising material, or comparing options — AI compresses the boring parts. The judgement is still yours.
Monday morning, 38 emails waiting. A Berlin marketing manager pastes the inbox into her enterprise Copilot and asks for a triage summary: which need a reply today, which can wait, which are noise. Twelve minutes saved. She still reads each one — but she reads them in priority order.
Two jobs it'll quietly get wrong
The other side of the deal. AI is bad at two things that matter at work, and the bad will sound exactly like the good — same calm sentences, same neat structure.
- Anything specific to your context that wasn't in the prompt. The unwritten rule that this client always wants the executive summary first. The fact that your CFO hates a certain corporate buzzword and will rewrite anything that uses it. The history of why this proposal failed in 2023. AI doesn't know any of that. It'll generate something plausible that's wrong in ways only a colleague would catch.
- Knowing when it's wrong. AI doesn't have a separate "I'm not sure" setting. It generates the most likely-sounding next sentence whether or not it's true. A made-up regulation, a misremembered date, a confidently-cited case that doesn't exist — they all arrive sounding identical to the real ones.
The career-protective skill is this: treat AI output the way you'd treat a fluent new starter on day three. They sound confident. They have absorbed a lot. They will also confidently get something important wrong, and you have to catch it before the document leaves the building.
A Toronto procurement analyst asks AI to draft a comparison of two cloud providers. The output is fluent and well-structured. It also confidently states one provider's contract clause that doesn't actually exist in the proposal. Caught at review. Quietly damaging if it hadn't been.
The skill that's gained the most value
If AI compresses the boring parts, what's left to be valuable about?
Judgement. The bit where you read what AI gave you and say "that's not quite right because…" The bit where you know which client wants the executive summary first. The bit where you sign off and own the answer.
The people doing well in 2026 aren't the ones avoiding AI. They aren't the ones handing it the wheel either. They're the ones who use it heavily and still own the answer. AI as a fluent collaborator. You as the one who knows when it's wrong.
Pick three tasks this week to try AI on — first drafts, summarising, comparing. Don't be precious. Read the output. Edit it. Decide what's yours.
A Singapore policy adviser tries AI for the first time on a routine task: summarising a 40-page regulatory submission into a one-page brief for her director. Twenty minutes instead of two hours. She reads the summary, fixes one wrong date, adds one paragraph the AI missed. If her company's rules don't require disclosure for routine internal drafting, no separate AI note is needed — the judgement was hers.
Try this for 10 minutes
Pick three tasks from this week's calendar that fit one of the three categories above:
- A first draft of something (email, slide, status report, agenda)
- A long thing to summarise (meeting transcript, document, email thread)
- A comparison (two reports, two proposals, two policies)
For each, write down: which AI tool is sanctioned for this at your company? If you don't know, pause this lesson and ask IT or your manager. The next lesson covers why that question matters.
Keep the list. You'll use it again at the end of the course.
Keep this
- Rule — AI is genuinely good at first drafts, summarising, and comparing. It's bad at your context, and bad at knowing when it's wrong. The judgement is yours.
- Phrase — "Treat it like a fluent new starter on day three." — sounds confident, gets things wrong, needs you to catch it.
- Don't — Don't hand AI a task and then forward the output without reading it carefully. That's how careers quietly get into trouble.
Pop quiz, no marks
-
A colleague says "AI is going to do my job." In one sentence, what's the honest reply?
Show answer
Probably not your whole job — probably some of your tasks. The practical move is to learn where AI helps, then keep the judgement yours.
-
AI gave you a fluent draft that sounds great. What's the one habit before you send it on?
Show answer
Read it for the things only you would know — the client's preferences, the unwritten rules, the dates and names. AI doesn't have those. You do.
Want to go deeper?
Each of these topics on Plain AI explores one idea from this lesson in more detail:
Sources
- Plain AI — AI and knowledge work [primary]
- Plain AI — "Will AI replace me?" [primary]
- Microsoft Work Trend Index — AI at Work (annual) [vendor]