One machine counts, the other one talks
AI for spreadsheets, no code · Lesson 1 / 22
Two different tools
A spreadsheet is a calculator with memory. It computes exactly what you wrote and never makes an arithmetic mistake. A language model is a machine that predicts plausible text. It is excellent at phrasing, explaining and turning intent into a formula, but it does arithmetic by eye.
That gives you the rule the whole course is built on: the sheet computes, the model phrases. The moment you ask a model to add up a pasted column, you get a number that looks like the truth. Looking like it is not the same as being it.
Where AI saves you hours
- Turning a task into a formula. You say "sum revenue by rep for April" and get SUMIFS with the arguments filled in.
- Reading someone else's formula. Paste a four-line monster and ask for a step-by-step breakdown.
- Text cleanup. Rules for normalising names, splitting an address into parts, fixing dates.
- Ideas for analysis. Ask what questions these data can answer and you will get cuts you had not considered.
- Draft conclusions. You supply the computed numbers, the model writes the report text around them.
Where it hurts you
- Arithmetic on pasted data. Totals, averages, percentages — the model invents them.
- Recomputing "how much did it come to". Even with the data in front of it, the answer is guessed, not calculated.
- References to ranges you never showed. The formula will be syntactically perfect and logically wrong.
- Functions your version does not have. You will be handed XLOOKUP while you are running Excel 2016.
Bad: "Here are 200 rows of sales. What is revenue by region?" Good: "Structure: A = date, B = region, C = amount. Give me a formula for the regional total, region in E2. Excel 2016, semicolon separator."
The second version gives you a tool the spreadsheet will run. The first gives you a pretty number that you personally will have to defend.
Insight. A model almost never says it is unsure about a total. Confident tone and factual accuracy are unrelated properties, and that gap is the single biggest trap in data work.
Common mistake. Asking the chat for the answer instead of the formula. The answer goes stale the moment somebody adds a row; a formula recalculates itself.
Pro tip. Build the habit: any model reply containing a number must become a formula in a cell. If it cannot be turned into one, you asked the wrong question.
Cheat sheet
- The sheet computes, the model phrases.
- Ask for the formula, not the result.
- Always state your version and function language.
- A number from chat is a draft; a number from a cell is a fact.
1. Why should you not ask a model to total pasted data?
2. Which request produces something useful?
3. Why name your Excel version in the request?