Writing
systems, databases, and what breaks in production
Articles
3 articles
- Part 1: What we learnt keeping PostgreSQL read models up to date
2026-09-06
The first part of Filtering at scale: why Audience Republic built materialized read tables over normalised PostgreSQL data, and what keeping them current involved.
- We still deploy JARs over SSH. An AI made it reasonable.
2026-09-05
Audience Republic still deployed Java JARs directly to 17 hosts over SSH. An AI safety layer made the old machinery visible, repeatable and much harder to misuse.
- Isolate the container, not the database
2026-08-04
Parallel AI coding agents broke my test suite with random database errors. The root cause was one hardcoded compose project name — and four wrong assumptions about what “isolated” means.