Software that has to keep standing.

Essays on architecture, modernization and delivery inside large organizations. Long-form, sourced, and unhurried: the systems in question took decades to build, so the essays can take a few weeks.

Start with the essaysSubscribe by RSS

October 2026

· Modernization · 4 min

The strangler fig, twenty years on

Martin Fowler named the incremental-replacement pattern in 2004. Two decades later it is quoted more than it is followed, and the gap is where modernizations die.

· Leadership · 3 min

When the internal platform goes on sale

Spotify opened a storefront for its internal engineering platform on October 8. An essay on what a platform team becomes when its customers start paying, and what Amazon learned about that in 2006.

· Delivery · 2 min

On the outage that came back ninety-six minutes later

GitHub went down twice on October 7, ten minutes each time. The second incident's own status update called it a reoccurrence, which is the word that separates a mitigation from a fix.

September 2026

· Architecture · 1 min

Everything ships a database

The most deployed software in the world has no server, no company, and no sales team. A short note on what SQLite's ubiquity says about infrastructure.

· Architecture · 6 min

Whatever happened to capacity planning

For forty years someone owned the forecast. Autoscaling did not eliminate the discipline; it moved it to the other side of the invoice. An essay on a job title the cloud dissolved.

· Modernization · 3 min

Y2K, reconsidered as a success

The disaster that did not happen is remembered as a panic that was wrong. The record says otherwise, and the difference matters for every deadline since.

· Practice · 3 min

On the colour of the bikeshed, sixty-nine years on

Parkinson's law of triviality is from 1957; the email that renamed it for software is from 1999. Both are quoted weekly and obeyed nowhere. On attention, consequence, and a budget for reviews.

August 2026

· Leadership · 2 min

The bus factor has a literature

Knowledge concentration can be measured from the repository itself, and researchers have been doing it for a decade. On the number nobody computes until the resignation.

The full archive →