Why boring tech wins in year three.

Every year has a new hot database. In 2020 it was CockroachDB. In 2022 it was Planetscale. In 2024 it was Neon. In 2026 it's whatever shows up at HackerNews this week.

We pick Postgres.

The boring-tech test

Before we adopt anything new, it has to pass three questions: 1. Will this still be maintained in 2030? 2. Can we hire engineers who already know it? 3. Is the failure mode well-documented?

Postgres passes all three. Most new tools fail at least one — often all three.

What "boring" actually means

Boring doesn't mean outdated. It means load-bearing, well-understood, battle-tested. Next.js is boring now. Postgres was boring in 2015 and is still boring. Redis is boring. AWS S3 is boring.

Boring tech has: - Documented edge cases - Recovery guides when it breaks - Hiring signal (LinkedIn searches) - Migration paths out, if needed - A community that's seen your problem before

The hidden cost of new tech

We've seen it twice: a YC-backed startup picks a 2-year-old vector database because the founders are AI-pilled. Two years later the vendor gets acquired and the product is sunset. Migration cost: six weeks. Migration risk: every one of their production features.

Meanwhile, a client we built on Postgres in 2018 is still shipping features on the same database today. Zero migration. Zero panic weekends.

When we break the rule

We use new tech when: - The problem genuinely has no boring solution (LLM inference in 2023) - The cost of switching later is low (swap a library, not a datastore) - The new tool is an incremental better of something we already use

So we use Claude + GPT for AI. We pick React and Next.js for frontend. We stick with Postgres for data, Redis for cache, S3 for files.

Founders thank us in year three.