Skip to content
experience

A normal day at a Japanese SIer, explained as politely as I can

Kazuhiro Yamamoto@yamamoto_kaz

June 24, 2026 · 3 min read

Good evening. Several members asked me in the comments what a "normal day" is like at a Japanese system integrator (we say SIer). I will try to explain honestly. I am sorry for my poor English and for the length. I wrote it on the train.

  1. 7:40 I take the train from my home in Sakai to the office in Umeda. It is crowded, but it is very punctual. The train is the most reliable system in my life, I think.
  2. 8:45 Chōrei (morning meeting). Our section stands up and one person gives a short speech. Last Monday it was my turn. I talked about hydrangeas, because it is the rainy season. Nobody asked about the project. This is normal.
  3. 9:00 to 12:00 I work on the migration of a regional bank's system from COBOL to Java. The COBOL was written in 1989. Many lines of the new Java look like the COBOL, but with semicolons. I am not proud of this, but the customer's requirement is "same behaviour, exactly", including two bugs that their branches depend on.
  4. 12:00 Lunch, bento at my desk. At 12:00 the lights on our floor are turned off to save electricity. We eat in the dark. Visitors find it strange. We do not notice any more.
  5. 13:00 to 17:30 Design documents. Our design documents are in Excel. Each cell is a small square, and we use the cells like graph paper to draw screens and tables. If there is a screenshot, we paste it into the Excel. One design document I maintain is 41 sheets and 38 MB. When I explain this to foreign engineers, they think I am joking. I am not joking.
  6. Approval When a document is finished, it needs approval stamps (hanko) from my leader, his leader, and the customer side leader. Now the stamp is often digital, an image of a red circle in the corner of the Excel. The image of the stamp is approved, but nobody reads sheet 27. I know this because there was a mistake on sheet 27 for eight months.
  7. 17:30 Official end of work. Nobody leaves. Leaving first is... how to say... a small courage.
  8. Wednesday Wednesday is "no overtime day". We are encouraged to leave at 17:30. Many people leave at 17:30, go to a cafe, and continue working on their laptop. I do not think this is what the HR department intended.
  9. Friday Sometimes nomikai (drinking party) with the section. Attendance is not required. Attendance is also not optional. I drink oolong tea and listen to my manager's stories about the 1990s, which are actually quite interesting.
  10. 21:30 Train home. I study English or read this website. Thank you for making my commute educational.

Some things I want to say so you do not misunderstand. My colleagues are kind and very careful people. Our systems run for decades because we are careful. The bank I work for has not had a major outage in my eleven years on the project. But I think sometimes we are careful about the wrong things, like the colour of the stamp, and not careful enough about the right things, like automated testing. Our regression test is an Excel with 2,300 rows that two people execute by hand for three weeks.

If I may ask one question to the members: in your company, who reads the design document after it is approved? I am curious if this is only a Japanese problem.

Thank you very much for reading.

Get the best of Dynamo Geeks weekly

One email each Monday with the most useful posts. No spam; unsubscribe in one click.

Comments

10

Sign in to join the conversation.

Sign in to comment
Sam Whitaker@segfault_sam
the image of the stamp is approved but nobody reads sheet 27 is the most enterprise sentence ever written 💀
6 net votes
Renee Whitfield@renee_em
This is wonderful, thank you. In US fintech the honest answer to your question is: the person who reads the design doc after approval is the new hire who joins 18 months later and finds out it describes a system that no longer exists. Is there any appetite on your team to move the regression sheet to automated tests, or is that a political question more than a technical one?
3 net votes
Kazuhiro Yamamoto@yamamoto_kaz
Thank you for your kind question. It is mostly political, I think. If tests are automated, the three weeks of manual testing disappear, and the three weeks are in the contract and in the invoice. I am sorry, it is a little sad to write.
0 net votes
Walt Hendricks@walt_mainframe
Mr. Yamamoto, the requirement to keep two bugs because the branches depend on them made me laugh out loud. We have three of those. One of them is from 1987. I would not dream of fixing it. Thank you for this.
3 net votes
Graham Whitfield@greybeard_graham
Mr Yamamoto, I have read a lot of posts on this site and I think this is my favourite. To answer your question: in thirty-four years, I believe I met two people who read the design document after it was approved, and one of them was an auditor. It is not only a Japanese problem. The Excel graph paper, however, might be. Cheers, G.
3 net votes
Kazuhiro Yamamoto@yamamoto_kaz
Thank you very much, Graham-san. Two people in thirty-four years is a very useful statistic. I will tell my leader (politely).
2 net votes
Dazza@dazza_rates_ur_desk
eating lunch in the dark at ur desk is the most depressing setup ive ever heard of. 2/10. the oolong tea gets u a point back
2 net votes
Ryan Goh@ai_maxi_ryan
bro just feed the COBOL into an LLM and get java out in an afternoon. 3 weeks of manual regression in 2026 is insane 🤖🚀
2 net votes
Kazuhiro Yamamoto@yamamoto_kaz
Thank you for your suggestion. If I may ask one question: when the generated Java has a different rounding for the interest calculation in one case out of 40,000, how would we find it without the 2,300 rows? I am sincerely asking.
0 net votes
Graham Whitfield@greybeard_graham
He won't answer that, Mr Yamamoto, but I enjoyed the question very much.
1 net votes