by Serguey Shinder
Through 2019 and most of 2020 my attention belonged to a new order platform we were building to replace the system we had run since 2014. I was in every design review, read every change, sat in the roadmap sessions and the architecture arguments, and enjoyed it more than anything else in those two years.
The 2014 system took ninety four per cent of our orders and effectively all of our money, and was looked after by two people and whatever I had left.
I found out what that cost on a Thursday in November, during a five hour incident in the old system about how a price band applied to part loads. I wrote a good deal of that code in 2015 and I could not help. I could not remember why the bands were expressed as they were, and one of the maintainers, who joined in 2021, explained my own design to me twice while a customer waited.
Afterwards I counted, which I recommend and did not enjoy. Of a hundred and forty code reviews I did that year, nine were in the old system. Of the recurring meetings in my calendar, four of thirty concerned it. The new platform carried six per cent of orders and was two years from parity.
The part I find least comfortable is that nothing had taken my attention away. It had no alerts, no escalations and no customers, and required nothing of me. I chose it every Monday, and it felt like the responsible choice, because it was the future of our stack, the thing I could describe at a conference, and the room where the interesting arguments happened.
What it cost was not the old system's health; two good engineers protected that. It was that the decisions with the most consequence were taken without the person who had most history of them, and twice that year I waved through an objection about the old system that I would have supported had I still understood it. The new platform, for its part, did not move faster for my presence. I was comfortably the fourth most useful person there.
So I swapped. The platform reviews went to an engineer who wanted them and had better instincts than I did, and I took the old system's roadmap, which nobody wanted and turned out to be where every question about our next three years actually lived.
The check I run now takes ten minutes a quarter. One list of where my hours went, one list of where the volume and the money are. If the two disagree by an order of magnitude I change what I am on rather than resolve to try harder.
Interest is not a signal about importance. It is a signal about interest, and in a week with nothing wrong in it, that had been the only signal I used.
– Serguey Asael Shinder
Leave a Reply