-
pgvector or a dedicated vector database? How a startup should choose
Choosing a vector database for your first AI feature: when pgvector in the Postgres you already run is enough, and the signals that it is time for a dedicated vector store.
Read post -
RAG not working? Chunking, retrieval and reranking mistakes to fix first
When a RAG feature gives vague or wrong answers, the model is rarely the problem. The chunking, retrieval and reranking mistakes to check first, in order.
Read post -
Do you need an AI agent, or a workflow with one smart step?
Agents are the most hyped way to build with LLMs and often the hardest to run in production. Here is how to decide whether your feature needs one.
Read post -
Adding an LLM without leaking customer data
Your first enterprise customer will ask where their data goes when your AI feature runs. Here is how to have a good answer before they ask.
Read post -
RAG or fine-tuning? How to teach an LLM about your business
Most startups that ask for fine-tuning need retrieval instead. Here is how to tell which one your feature needs, and what to build first.
Read post -
What "boring" engineering actually means
Boring is not slow, simple, or unambitious. Boring is the dial we turn up when reliability matters more than novelty — which is most of the time.
Read post -
Stripe Connect: the playbook we wish we had
Five years of shipping Stripe Connect — marketplaces, payouts, dispute handling, and the edges that turn a working demo into a robust production system.
Read post -
Why we reach for Postgres first (and second)
Most products do not need a new database. They need someone who knows Postgres well enough to stop reaching for new ones. Here is our default, and the few times we break it.
Read post -
The deploy pipeline that does not page you at 3am
A good deploy pipeline is boring on purpose: small changes, fast rollbacks, and enough observability to know what broke before a customer tells you.
Read post -
Shipping AI features without betting the company on them
AI features are easy to demo and hard to make reliable. The boring approach treats the model as one unreliable dependency inside a system you still control.
Read post -
An MVP is not a disposable prototype
The "minimum" in MVP is about scope, not quality. The parts you cut and the parts you never cut are what separate a fast launch from a slow rewrite.
Read post