AI at work · a plain-English guide

AI, made simple —
from asking to doing

No experience needed. We’ll follow one real job all the way — from typing a question, to AI that does the whole task for you.

An engineer reviewing protection test reports at a desk with a helpful digital assistant nearby
For engineers ~8 min read 🔒 Work data → Microsoft Copilot only
Scroll to begin ↓
You’ve already started

You already know how to use AI

Ever typed a question into a chat box and read the answer back? That’s it — you’ve already used AI.

💬Think of it as texting a very fast helper who has read almost everything. You ask in plain words — it replies in seconds.
The catch: on its own, you still do every step by hand — ask, copy the answer, then go do the next thing. This guide shows how to hand more of that work over, safely.
One example, all the way through

Meet the job we’ll follow

To keep it concrete, we’ll use one task that comes back every month — and watch how AI can take on more and more of it:

The monthly job

The protection-test report summary

Every month a batch of relay test reports lands. Someone has to pull the measured trip time out of each one, check it against the setting, flag any that miss, and email the team a tidy summary.

🧑‍💼And one picture to hold onto: imagine a new assistant just joined your team. How much you let them do — and how much you check — is exactly how AI “grows up” over the next four steps.
The big picture

AI grew up in 4 easy steps

Each step keeps the easy parts and hands off more of the boring, repetitive ones:

1

Ask clearly

Say exactly what you want. (Prompting)

2022–24
2

Show it the real stuff

Give it the actual files, not a description. (Context)

2025
3

Let it do the job

Give a goal; it works through the steps. (Agents)

2025–26
4

Set it on a loop

It runs the routine again and again; you supervise. (Loops)

Now · 2026
Step 1 of 4 · Ask clearly

Tell it exactly what you want

The clearer you ask, the better the answer. Same as briefing a new assistant on day one.

Giving a clear, written brief to a new assistant
Day one: a clear, specific brief beats a vague one.

Our job, step 1: you just want the summary email drafted. Watch the difference:

Too vague
“Write an email about the testing.”
Clear & specific
“Write a short, friendly email to the team. List which relays were tested this month, which ones missed their trip-time setting, and what the next step is for each.”
Good recipe: who it’s for · what’s going on · do this · like this · not that.
Step 2 of 4 · Show it the real stuff

Show it — don’t just describe it

A clear ask only goes so far if the AI can’t see what you’re talking about. So hand it the real thing.

📂Like your new assistant: saying “the trip times look off” helps no one. Handing over the actual test reports does.
Handing over the real reports and files to the assistant
Give it the real reports, one good example, and the rules.

Our job, step 2: instead of describing the reports, you attach them:

📄 Attach the actual test reports  ·  ✅ Show one summary you liked before  ·  🚫 Say the rules: “flag anything slower than its setting.”
Step 1 asks “what do I say?” Step 2 asks “what can it see?”
Step 3 of 4 · Let it do the job

Give a goal — it does the steps

Now you stop doing each step yourself. You give the goal, and it works through the whole task — then you check.

The assistant working through the whole stack of reports on its own
You give the goal; it reads every report and drafts the summary.

Our job, step 3 — same task, two ways:

😮‍💨 By hand

  1. Open report 1, find the trip time, type it in a sheet.
  2. Compare it to the setting by eye.
  3. Repeat 14 more times.
  4. Spot the misses yourself.
  5. Write the email yourself.
~30–40 min · easy to miss one

🚀 Give it the goal

  1. Point it at the folder and state the goal.
  2. It reads all the reports.
  3. It builds the table and flags the misses.
  4. It drafts the team email.
  5. You check and send.
a few minutes · you check, not retype
Same AI underneath. The difference is how much it’s allowed to do — and you always check before anything goes out.
Step 4 of 4 · Set it on a loop · the newest one (2026)

Set it up once — it keeps going

The job comes back every month. So instead of re-asking each time, you set up the routine once — and it runs itself. You just supervise.

🔁Your assistant now owns the monthly report. When the new reports land, they just do it — and drop the draft on your desk to approve.
A monthly routine running on its own while a person supervises
Set the routine once; it repeats each month with you watching.
New reports land→ It builds the summary→ It flags the misses→ You approve↺ next month
This is the brand-new one everyone’s talking about. At work you’ll meet it through Copilot’s agent mode — always with a person watching.
Names you’ll hear

The famous tools, in one line each

ToolMade byKnown for
Microsoft Copilotour tool at workMicrosoftBuilt into Word, Excel, Outlook & Teams — works on your own files.
ChatGPTOpenAIThe famous all-rounder for everyday questions and writing.
ClaudeAnthropicStrong with long documents and careful, detailed answers.
GeminiGoogleTied into Google’s apps and search.
✅

The easy rule for us

At our firm we use Microsoft Copilot — it’s the only one cleared for company and client information. The rest are just good to know by name.

The good news

Ours is already in your apps

Nothing to install. It’s built into the apps you already use every day:

Word

Write, shorten, or summarise a document.

Excel

Make sense of a table and chart it.

Outlook

Summarise long chains; draft replies.

Teams

Catch up on a meeting; list the to-dos.

PowerPoint

Turn a document into a first draft deck.

VS Code

For coders: write and explain code.

What you see depends on your licence — if you don’t see it in an app, ask IT.

Try it today

Three steps you can take, easiest first

All three are the same monthly job — just handing over a little more each time.

① Warm-up · Excel

Make sense of the numbers

  1. Turn your readings into a table: Insert ▸ Table.
  2. Open Copilot, then ask in plain words:
  3. “Total each panel’s connected load and chart the 10 biggest.”
  4. Check what it made, then drop it in the sheet.
② Step it up · Outlook & Word

Tame a long email chain

  1. Open that 30-reply test-schedule chain.
  2. Ask: “List every commitment, who owns it, and its due date.”
  3. Then: “Draft a status email — and a short note on what’s slipping.”
  4. Read it, make it yours, then send.
Copilot — Outlook
List every commitment, owner, and due date from this chain.
7 found — 2 are past due: the relay settings and the sign-off. Want a status email plus a slip note?
Yes — both, please.
③ Go big · agent mode

Let it run the whole summary

  1. Turn on agent mode where it’s offered.
  2. Give a goal, not one instruction: “From these reports and the load list, build a one-page summary, cross-check the key numbers, and flag anything that doesn’t match.”
  3. Watch it read each file, draft, and cross-check — step by step.
  4. You’re the checker: confirm the flags and facts before you use it.

This one is new and changing fast — check what’s switched on for you before you rely on it.

A few simple rules

Stay safe, stay in charge

An engineer reviewing and approving the finished work, in control
It drafts; you decide. Your name is on the work.
🔍

Always check it

It can sound sure and still be wrong. Check the numbers and names yourself.

👷

You’re still the boss

It’s your judgment and your name on the work — not the AI’s.

🔒

Company info → Copilot only

Never paste company or client information into any other AI tool.

📝

It’s a draft, not the finished job

A fast first draft. You finish it.

Steal these on Monday

Handy things to say

Sum this up and list what’s still not decided.
Rewrite this for a client — friendly, 150 words.
Turn these notes into a to-do list with names.
Explain this simply, then give the detailed version.
Check this and tell me what you’d double-check.
Give me 3 options, with the pros and cons of each.
Your turn

Try the one that earns its keep

Don’t stop at summaries. Point it at the job that normally eats an afternoon — checking the test results against the design:

The cross-check that used to take an afternoon

Catch where the test results no longer match the design

Give it the test reports + setting schedule + coordination study→ “Find every relay whose measured trip time no longer matches the study”→ “List which downstream feeders could lose discrimination”→ “Rank them by risk and show your working”→ You review & decide

That’s three different documents, cross-referenced line by line — the kind of mismatch that’s easy to miss by hand, surfaced in minutes. This is where it stops being a chatbot and starts being a second set of eyes.

The whole idea in one line: ask clearly, show it the real stuff, let it do the heavy cross-checking — while you stay in charge and make the call.

Questions? 🙌