You Don't Have to Be an Engineer. How to Dump “That One Chore” onto AI, by Job Role

Friday evening, the sales guy at the next desk let out a sigh.

“Thirty business cards this week. Typing them all into Excel before I go home — isn’t this brutal?”

I leaned over, looked at his screen, and said: “You don’t have to type those, you know.”

I still get the feeling a lot of people think AI is a tool for engineers to write hard programs with. But where it actually earns its keep is somewhere far more humble: that annoying manual chore you repeat every single day. Transcribing business cards, sorting inquiry emails, cleaning up meeting minutes, churning out product descriptions. This kind of “nobody wants to do it, but somebody has to” work is exactly what AI is best at.

Today I’ll go role by role and talk about handing off “that one chore” to AI. There’s barely any code. Where there is, it’s the kind you copy-paste and rename — nothing more.

So what can a non-engineer actually do?

Let me clear up the misunderstanding first. You do not need programming knowledge to give AI a job.

Think of it as an incredibly quick-learning new hire. Say “here, handle this” in plain English and it’ll mostly get it done — building an Excel formula, polishing prose, lining up scattered data into a table. The one catch: it’s a new hire, so if you don’t tell it “what, how far, and in what shape” up front, it’ll come back with the wrong thing. That’s the only knack.

This article assumes two tools. One is the ChatGPT or Claude chat window — anyone can use it right away. The other is Claude Code, a kind of AI that can read and write files directly. The name has “Code” in it, which makes people brace themselves, but think of it as the clerk that “looks at the files in a folder, tidies them, and saves them.” Not a coding specialist — a stand-in for your computer chores.

The official write-up is in the Anthropic docs. It’s in English, but just glancing at the vibe is reassuring.

From here, let’s go role by role and look at “where it pays off.”

Case 1: Sales — escaping the transcription hell of cards and emails

Back to the business cards. If you work in sales, transcribing cards and email signatures into a customer list follows you everywhere. Two minutes a card sounds tiny, but 30 cards is an hour. It quietly steals your evening.

You can end this just by asking the AI to “turn this text into a table row.” It works whether the text came from a card-scanner app or you pasted the signature from the bottom of an email. You’re asking it to turn a messy blob into one clean row of a table.

For example, paste this into the chat:

From the text below, extract company / name / department / title / email / phone,
and make a single CSV line. No header row. Leave a field blank if it's missing.

Sample Trading Co., Sales Dept., Manager
Taro Yamada
[email protected] / 03-1234-5678

What comes back is one line: Sample Trading Co.,Taro Yamada,Sales Dept.,Manager,[email protected],03-1234-5678. Paste that into a spreadsheet and you’re done. Paste ten cards’ worth and ten rows come out at once.

Take it one step further and you can hand off the email draft too. Ask for “a thank-you email to this person for the trade show the other day — polite but not stiff” and you’ve got a starting draft in three seconds. Writing from zero versus fixing up a draft that’s already there — the time difference is night and day. There’s a fuller take on sales email automation in email automation with Claude Code.

Case 2: Customer support — get ahead of the inquiry sorting

A support person’s morning starts by opening the piled-up inquiries and sorting them — “this is a refund,” “this is a bug report,” “this is a sales lead.” Reading and classifying them one by one quietly drains the brain, doesn’t it?

This “read and classify” is something AI is genuinely great at. Hand it the inquiry text and have it return the “type,” the “urgency,” and a “draft first reply” all at once. The support rep just checks the sorting that came out and polishes the response.

But here’s the part that matters most in support. Never let the AI send its reply straight to the customer. It’s a draft, always. A human reads it last and sends it. Hold that one line (the failure story later shows you why, painfully).

Show the AI a handful of “common past questions and answers” and it’ll write replies that lean toward your house tone. Even a fine-grained rule like “we don’t say customers, we say members” gets honored if you tell it up front.

Case 3: Admin — merging 30 Excel files into one

The thing I hear most from admin folks is “merging multiple files.” Thirty CSVs or Excels scattered by branch and by month, and you have to merge them into one table and total it up. By hand it’s a blizzard of copy-paste, and one row out of place throws the numbers off.

This is Claude Code’s moment. Hand over the whole folder and ask, in plain English, “read every CSV in here, merge them into one, and give me the per-branch totals.” Whether it’s 30 files or 300, the wait is about the same.

“But running it on my own PC sounds hard,” you might think. Just to try it, you can have the chat AI build you a single-file script like this. Below is a dead-simple example that only “stacks every CSV in the same folder into one sheet.” If you have Python installed, save it as merge.py and it runs with what amounts to a double-click.

# Merge every CSV in the same folder into one sheet (cure for admin copy-paste hell)
# Usage: drop this file in the same folder as the CSVs and run it
import csv, glob, os

rows = []
header = None
for path in glob.glob("*.csv"):
    if os.path.basename(path) == "merged.csv":
        continue  # skip the output file itself
    with open(path, encoding="utf-8-sig", newline="") as f:
        reader = csv.reader(f)
        head = next(reader, None)  # treat the first row as the header
        if header is None:
            header = head
        for row in reader:
            rows.append(row)

with open("merged.csv", "w", encoding="utf-8-sig", newline="") as f:
    writer = csv.writer(f)
    if header:
        writer.writerow(header)
    writer.writerows(rows)

print(f"Wrote {len(rows)} rows to merged.csv")

It’s fine if you don’t follow every line. Basically it’s the clerk that “stacks the folder’s CSVs vertically into one sheet.” When you want to go further into totals and conditional logic, automating spreadsheet workflows with Claude Code has the practical steps.

Case 4: Marketing and writers — the guard against duplication and inconsistent terms

Last up, marketing and writers. Twenty product descriptions, a week of social posts, ten ad copy variants. “Producing volume” is a perfect match for AI.

But the common failure here is “I let the AI write all of it and everything came out sounding the same.” The knack: decide the template and the raw material yourself, and let the AI do the fleshing-out. Hand over just “the product’s three features,” “target is dual-income thirty-somethings,” “here are the banned words” — and have it mass-produce combinations of phrasing. The human moves into the role of picking from what comes out.

The other quietly powerful one is unifying inconsistent terms. “Web / web,” “e-mail / email,” “log-in / login.” Paste the draft and say “our style rule is this — fix it,” and it’ll catch even the inconsistencies your eyes skip over. That last bit of proofreading polish gets a lot lighter.

Three accidents I caused myself

Let me be honest. Early on, I face-planted pretty spectacularly.

One. I almost let the AI auto-send support replies. “The drafts are good enough now, can I just send them?” — a moment of weakness. Mid-test, the AI misread one refund condition and drafted “we will issue a refund” for a case that should have been declined. If I’d automated the send button, it would have gone straight to the customer. Keep the send in human hands, always. That one sank into my bones.

Two. My instructions were too vague. On an admin merge, I said only “merge this nicely,” and the totaling method and sort order were left to the AI’s mood — a different table came out every time. The moment I spelled out the finished form — “sort by branch name A→Z, total on the bottom row” — it locked in. Just handing over what shape you want it in makes it a different beast.

Three. I had it mass-produce without raw material. I dumped “write this product description attractively” and every product came out with empty, one-size-fits-all lines like “high quality and easy to use.” Once I handed over the spec sheet and the target persona, each product finally showed its own face. The AI is a cook who works from ingredients, not the one who invents the ingredients. I had that backwards.

If you’re starting out, start here

Don’t aim for “let AI do everything” right away. You’ll bounce off.

What you pick is just one annoying chore you do every week. Card transcription, inquiry sorting, file merging, term unification. Narrow it to one. Then proceed like this:

1. Write what you want done as if explaining it to a new hire (context, material, the shape you want)
2. The very first time, always check the result yourself (don’t trust it blind)
3. Keep “send / delete / money moves” operations in your own hands

Hold those three and no big accident happens. Start with one email’s worth, one card’s worth, today.

What actually happened when I tried it

That sales guy from the intro? After that, he converted his 30 cards to text, built the customer list in five minutes, and went home. “I had no idea clocking out could be this early,” he laughed. I won’t forget that.

For myself, after delegating one chore per role, the three where the effect was unmistakable were transcription, sorting, and merging. The common thread: work that barely uses your head but devours time and effort. Conversely, replies that read someone’s emotions and brainstorming campaign ideas — even with an AI draft, I ended up heavily rewriting them. That part stays human work.

So here’s my conclusion. Hand the AI the “annoying but brainless” work; keep the “judgment and the final push” for yourself. Draw that one line and, regardless of role, you’ll reliably shave one or two hours of overtime off your day.

Wrapping up

AI isn’t a tool just for engineers. Sales transcription, support sorting, admin merging, marketing volume — every role has “that annoying chore,” and most of them can be handed off.

Only three things matter. Start with one chore / state the shape you want concretely / keep “send, delete, money” in your own hands. No hard programming required. Try it with one business card you got today, one inquiry from the pile.

When you want to map out “where in our work can we hand things to AI” across a team or department, take a peek at the training and consultation page too.