by Serguey Shinder
The slide said Platform Team, and underneath it were the reasons. Tooling had diverged across the squads. Nobody owned upgrades. Three teams had built three different deployment scripts. Reliability work only happened after an incident. Every point on that slide was true, and I had sat in the meeting before, running in the other direction.
Seven years earlier, in the same building, there had been a central infrastructure team and a slide about breaking it up. The reasons on that one were also all true. The central team was a bottleneck. Tickets sat for a fortnight. They did not understand the product context and made decisions that were technically defensible and commercially wrong. Squads should own what they run.
So we decentralised, the diseases of decentralisation duly arrived, and now here was the cure: the structure whose diseases I could describe from memory, having lived through them the first time.
I am not being cynical about it. Centralising was probably right for where the company was. Both structures work and both rot, in opposite directions, and an organisation of any age swings between them on roughly the timescale it takes for the people who remember the last swing to leave. In my experience that is about five years, roughly how long it takes for the pain of a structure to outgrow the memory of the previous one.
What struck me in that room was not the decision. It was that nobody presenting it knew the earlier decision had happened. It was not hidden. It had simply never been written anywhere a newcomer would find it, so the reasoning left with the people, and a swing back was being experienced as a discovery. That is the expensive part. Not the oscillation, which may be unavoidable, but going round the loop with no memory of the last lap and therefore no possibility of taking a shorter one.
I said one thing, and it was not that we should stop. I asked that we write down what we were giving up rather than what we were gaining, and what we would expect to see if we had overshot. Two paragraphs. If tickets to platform begin sitting for two weeks, that is the old disease returning, and somebody should recognise it rather than discover it.
People ask what skills will still matter in ten years and the answers are usually about technology. The one I would actually bet on is duller than that. Every organisation needs somebody who remembers why the last decision was made and can say so without sounding like an obstacle. There are fewer of them than there used to be, because almost nobody stays long enough, and the documents that could stand in for them are almost never written.
I have stopped arguing about which structure is correct. I argue about whether anyone recorded why we chose it.
– Serguey Asael Shinder
Leave a Reply