The Database Marathon started on a bike ride. Funny, given the name.
A few weeks ago, out on the bike early on a Saturday morning, I was turning over the week in my head — work included. And it struck me that almost nobody sees the whole of it. Pieces show up in dailies, pieces in specifications, but the end-to-end picture is rare — apart from our Team Leader, who I've spent hours discussing the technical details with.
That matters more than it sounds. From the outside, months of work can look like a handful of unrelated tweaks — or like someone found a silver bullet. It was neither: it's a long chain of connected decisions, each making the next possible.
I thought about writing it up as internal documentation. But a wiki page gets read out of obligation, if at all. An article, though — an article someone might actually read, maybe even outside the team.
So I came home, and my family kindly let me disappear into the study for the day. (After second breakfast — the privilege of training early on a Saturday.)
Then I just talked. Two and a half hours into my phone: everything I could remember, out of order, with some numbers and names left as gaps.
Afterwards I cleaned up that transcript myself — sorting it, filling the gaps, fixing what came out scrambled — and handed it to Claude to help turn into an article. It suggested straight away that this wasn't one article. It was five.
From there we worked like an author and an editor: I decided what the series should say, and Claude pushed back on structure, checked facts, and caught the places where I'd been vague. It took three or four days, spread over a weekend and several evenings: edits, line-by-line proofreading, technical fact-checking, digging out details I hadn't remembered, and working out where to publish.
So, plainly: this series is based on my own experience, written with AI as a tool.
Which brings me to what nearly stopped me from publishing. The feed is increasingly full of text that reads well and says very little — AI slop, as it's known — and the accounts written by people who actually did the work get lost in it. That was exactly my worry: publishing something written with AI help and being taken for part of that noise.
The tool isn't what makes the difference. What matters is whether there's anything underneath it — experience, decisions, something at stake. Behind these five parts are months of real work; AI helped me write it down and keep it accurate. With it, the whole thing took a few days. Without it, it would have taken zero hours — because I would never have started.
And if AI can fill a feed with noise, I'd like to believe it can help clear it out as well.
Thanks to the colleagues and friends who read the early versions, pushed back, and pointed me to other places to publish. And thanks to everyone who read the series — I hope it was useful, or at least interesting.
Originally published as a LinkedIn post, alongside the series "The Database Marathon."
Top comments (0)