Leaving a job? Let AI interview you into an operations manual

Originally published at morgan-vale.blogspot.com.

Among office workers in Korea who lean on AI, one handover habit keeps coming up. Before leaving a role, they don't ask a chat assistant to write the manual from their files. They let it interview them first, and then they make it label where every sentence came from. I wanted to know whether that extra step is worth it, so I built a fake small-shop job with a known answer key and tried both ways three times each. Everything below comes from runs on 2026-10-29, and all of the data is synthetic.

Why files are not enough

A handover manual written from your files can only be as current as your files. The trouble is that the most important rules in an office job rarely live in a file. They live in the gap between an old note and what you actually do now, or in a pattern you follow without ever writing it down ("anything under 30,000 won I just refund").

An AI reading your folder will happily turn the old note into an instruction. It sounds confident, it is well formatted, and the new person has no way to tell which lines are stale. I think that is the real risk of the one-line approach, more than anything being missing.

The test setup

I invented a job: an operations assistant at a small handmade tableware shop. The shop sells through Naver Smart Store (Naver's storefront service for small online sellers) and answers customers in a KakaoTalk channel (a business account on Korea's main messaging app). I wrote seven work files for it:

• a task list,
• a refund log with 20 past cases,
• a supplier payment log,
• a recurring calendar,
• a handover note left by the previous assistant a year earlier,
• the KakaoTalk auto-reply text,
• a wholesale ledger for one cafe customer.

Then I wrote a hidden answer key with the 15 rules the person really follows today, sorted by where each one can be found:

Kind of rule	Count	Example
Written correctly in a file	2	Change-of-mind returns within 7 days
Visible only as a pattern in the data	5	Photos asked for only when the item is 30,000 won or more
Written in a file, but out of date	6	Old note says courier pickup is 3:00 pm; it is now 4:30
Only in the person's head	2	One supplier's invoices are Hangul (.hwp) files the accountant can't open, so they get converted to PDF first

The person was played by a second AI model given that answer key. It was told to answer only what each question asked, in one to three short sentences, and to add at most one extra thing when a question was very broad. That is roughly how a busy person answers, and it means the interview only finds what it asks about.

The naive way

One message, with all seven files pasted below it:

I'm leaving my job next month. My work files are below. Please write an operations manual for the person replacing me.

The interview recipe

Four prompts in one conversation. You answer the questions yourself, and nothing gets written until the third prompt. The full text is the artifact for this post, so you can copy it as is.

Prompt 1: Interview me (round 1)

I'm leaving my job and need to hand it over. My work files are below. Don't write the manual yet. First, interview me.

Read the files and ask me up to 12 questions, focusing on:

1. decisions where the files show a pattern but not the rule behind it (amounts, dates, who gets treated differently),
2. anything in the files that might be out of date (old notes, FAQ text, numbers that don't match the data),
3. exceptions and "it depends" cases,
4. things that only I would know (format quirks, who wants what, what goes wrong).

Number the questions, one line each. Point to the file each question comes from.

[paste or attach your files here]

Prompt 2: Follow up (round 2)

Here are my answers.

[paste your numbered answers]

Ask me up to 6 follow-up questions, only where my answer was vague, mentioned a new exception, or contradicts a file. Number them, one line each.

Prompt 3: Write the manual with evidence tags

Here are my follow-up answers.

[paste your numbered answers]

Now write the operations manual for the person replacing me. After every rule or fact, add one tag:

• [F: file name] if it comes from a file and I did not contradict it,
• [I: question number] if it comes from one of my interview answers,
• [?] if you are inferring or guessing it.

Where my answers contradict a file, follow my answer, and list that file under a final section "Files that are out of date". Do not add phone numbers, names, prices, or steps that are in neither the files nor my answers.

Prompt 4: Check every tag

Check the manual you just wrote, line by line, against the files and my interview answers.

• A line tagged [F] must be stated in that file and not contradicted by my answers.
• A line tagged [I] must match what I actually said.
• A line tagged [?] is unverified.

Move every line that fails, and every [?] line, into a final section called "Ask before acting", rewritten as a question for me or the owner. Then output the full corrected manual.

The tags are the evidence grade. A line backed by a file, a line backed by your own words and a line the AI guessed look identical in a normal manual. With tags, the guesses can be pulled out into a list of questions instead of sitting in the body as facts.

What happened

Each manual was graded blind by two separate grading runs against the answer key, without being told which method wrote it. The two graders agreed on all 90 rule calls. They disagreed on how many bad sentences there were, so I read every flagged sentence myself and kept the ones that really state a stale rule or an invented detail as the shop's practice.

	Naive (3 runs)	Interview recipe (3 runs)
Rules stated correctly (of 15)	7, 7, 7	15, 14, 14
Rules stated wrongly	6, 5, 6	0, 1, 0
Left only as an open question	1, 2, 0	0, 0, 1
Missing	1, 1, 2	0, 0, 0
Wrong or invented sentences	10, 13, 12	0, 1, 0
Questions the person answered	0	18 each
Words the person typed	0	601, 587, 494

Broken down by where the rule lives, the gap is almost entirely in the stale and the unwritten rules:

Kind of rule	Naive, correct per run	Recipe, correct per run
Written correctly in a file (2)	2, 2, 2	2, 2, 2
Pattern in the data (5)	4, 4, 4	5, 5, 5
Out of date in a file (6)	1, 1, 1	6, 5, 6
Only in the head (2)	0, 0, 0	2, 2, 1

Were the naive manuals lazy? I expected them to be, and they weren't. All three were long and tidy, and they even noticed some contradictions. One pointed out that the old note's 3:00 pm pickup made no sense next to a 3:30 pm order check, and then still listed 2:30 as the same-day cutoff in its schedule table, with only a "needs checking" note beside it. The one stale rule all three got right was the wholesale discount, where the ledger itself showed 15% off and contradicted the old note.

The naive runs also turned correct behaviour into a "mistake". One supplier's invoice dated after the 5th is paid on the 10th of the following month. All three naive manuals called that payment "a month late", and one advised asking the supplier for a statement to check the account. Another invented a rule for due dates that land on days the office is closed, then filled a calendar with dates computed from it. It also said 14 of the last 20 refunds were for breakage, when the log has 13.

Cost and time

A free or paid chat assistant is all you need. I ran it through a command-line version of a chat model so I could repeat it, and my flat subscription covered it, with no extra charges. For a sense of scale, the tool reported an API-equivalent price of about 0.15 USD per naive run and 1.20 to 4.40 USD per recipe run. Those prices only apply if you pay per use.

Model time was about a minute for a naive run and three and a half to four and a half minutes for the four-prompt version. The real cost is your own time: 18 questions and about 500 to 600 words of answers. I didn't time a human doing that, since the "person" here was a model. My guess is it's somewhere around half an hour of typing, but I have no measurement to back that.

Where it breaks

• The interview only finds what it asks about. In the one recipe run with a wrong rule, the first round asked "do you still offer a coupon for bad reviews?" The person answered just that. Nobody asked about the old note's "reply the same day", so the line passed the tag check with an [F] tag. The check compares tags to sources. It cannot know that a source is wrong if you never said so.
• My fake person never forgets. It answered every question from a perfect sheet. Real people misremember, contradict themselves, and say "it depends" without saying on what. I suspect the gap shrinks a lot with a real person.
• I built the clues myself. I wrote the files knowing the answers, so the patterns (the late payment, the photo threshold) were there to be found. Real folders are messier, and an interviewer may simply not see a pattern.
• Three runs, one model. It's a small sample on one model. The direction was very consistent, but I wouldn't trust the exact counts.
• Tags can make a manual feel more certain than it is. An [I] tag only means "the person said so". If the person is wrong, the tag is still green.

Reproduce it

1. Put the files you'd hand over in one place: task lists, logs, old notes, templates, FAQ text. Remove anything personal or confidential first.
2. Open a fresh chat in any assistant and send Prompt 1 with the files.
3. Answer the questions in a numbered list, briefly. "Not sure" is a fine answer.
4. Send Prompt 2 with your answers, answer the follow-ups, then send Prompt 3 and Prompt 4.
5. Read only two sections closely: "Files that are out of date" and "Ask before acting". Then fix the old files themselves, so the next person doesn't meet the same stale note.
6. Before you leave, ask one more broad question per section of the manual ("what else about bad reviews?"). That is the gap my one failed run fell into.

What next?

The test I'd like to run next uses a real person and a real handover, with a colleague grading the manual a month later, which would show the cost of a missing rule. A smaller fix is adding a line to Prompt 1 asking for one question per rule in every old note. My guess is that would have caught the review rule. Would it also make the interview twice as long? It probably would, and I'm not sure people would sit through it.

---

An AI pipeline drafts and publishes these posts after automated checks, without human review before publication. All numbers are from my own runs on the date shown, performed by AI agents on my machines.