390 saved links, 2017–2026 (peak: 2024). Two collections wearing one tag: the timeless shelf — CMU lectures, B-tree explainers, Postgres-scaling war stories — and the AI-era overlay that exploded in 2024, when every database grew a vector column and every LLM learned SQL. Your 2026 saves mark the third act: agents as first-class database clients, and the question of what happens to a system designed for predictable queries when the queries stop being predictable.
Related: Data Engineering · RAG · Azure · System Design & Architecture · SRE & Observability
How databases actually work — the canon
- CMU 15-445: Intro to Database Systems — “stop memorizing DB concepts, start understanding how databases actually work”; the save calls it pure gold, and it is.
- CMU 15-721: Advanced Database Systems — the sequel, for after 15-445.
- MIT 6.824: Distributed Systems — the reading list and lectures, the natural next door for a backend/infra person.
- Things I Wished More Developers Knew About Databases (rakyll) — years old, still the best checklist of hard-won truths.
- B-trees and database indexes — “the best explainer on b-trees that you are going to get.”
- Phil Eaton: a write-ahead log is not a universal part of durability — internals in a highly digestible format.
- Asianometry: The Birth of SQL & the Relational Database — the history lesson, fish pun included.
The agent era — when AI becomes the database client
The freshest thread in the tag, and the one closest to where you’re heading.
- The contract shift: Arpit Bhayani: defensive databases — “have agents broken the unspoken contract we had with our databases?” · Databricks: how agentic software development will change databases — every service will grow these capabilities.
- Agent-ready guardrails: Supabase’s postgres-best-practices — Agent Skills for performance, security, and schema design, react-best-practices style · Seroter: one prompt to design a multi-tenant SaaS billing schema — “so easy, it’s almost embarrassing.”
- MCP plumbing: the Spanner MCP server codelab · MCP Toolbox for Databases · Gen AI Toolbox for Databases — Google’s open-source path from agent to production database.
- NL2SQL, done carefully: a multi-agent SQL assistant with human-in-loop checkpoints and cost control — trust as a design feature · embedding trust into text-to-SQL agents · the LangGraph SQL agent tutorial.
- NL2SQL, everywhere: plain-English queries over Cosmos DB with MCP + Semantic Kernel · an analytics agent on LangChain + DuckDB, no SQL written · an LLM SQL agent for SAP HANA · natural-language Postgres — try it yourself · SkyRL-SQL: RL-training a text-to-SQL model · LLMs are getting better at SQL.
Vector databases — the 2024 gold rush
The peak-year story: embeddings arrived and the database world reorganized around them.
- Vector databases explained in 3 levels of difficulty — Leonie Monigatti’s classic; you saved the 2026 remix too.
- calculating a vector database by hand — the author made it to show his son that high-school cosine “is what powers vector search”; part of the by-hand RAG / vector DB / agents slide series.
- Cassie Kozyrkov: embeddings, vector search, k-NN, ANN — the vocabulary in one pass.
- Weaviate: what is a vector database? · the Getting Started with Vector Databases refcard · vector databases in AI and LLM use cases.
- Pinecone’s RAG pipeline design questionnaire — storage, embedding models, chunking, personalized.
- Information Retrieval flashcards — 37 things learned in two years at a vector database company, as spaced repetition.
Postgres & the SQL workbench
- lessons from 5 years of scaling PostgreSQL — bloat, upgrades, XID wraparound (“do you monitor this?”).
- a RAG app in Python with just Postgres and pgvector · even simpler with pg_vectorize · an autonomous agent on pgvector + LangChain — the “Postgres is your vector database” caucus.
- postgres.new — in-browser Postgres with AI schema help and postgres.ai — ask it to visualize your DB schema.
- Datadog’s free SQL execution-plan visualizer — paste a plan, spot the missing index · Database Monitoring index recommendations — the observability angle you already speak.
- drawDB — diagram editor and SQL generator (the repo).
- Advent of SQL — free challenges with in-browser playgrounds · advanced SQL for data science.
- The SQLite corner, via Simon Willison: SQLite as a production Rails backend and Cloudflare’s SQLite-backed Durable Objects — thousands of tiny databases as an architecture.
The Azure cluster — Cosmos DB and Azure SQL go AI
Your platform depth, applied — the densest vendor thread in the tag.
- building AI agents with vector search in Cosmos DB · the Cosmos DB + Azure OpenAI Python developer guide (DiskANN) · the RAG chatbot quickstart notebook · Cosmos DB Python getting-started.
- native vector support in Azure SQL + Azure OpenAI · the Azure SQL “cryptozoology” AI embeddings lab — vector search practice with monster data.
- choosing the right Azure vector database — the decision guide for the overwhelming landscape.
- Vectors, AI, Agents: how AI changes database interaction — agents generating T-SQL with reasoning and guardrails.
- 10 free Azure SQL databases per subscription — the zero-cost lab bench.
- the VS Code SQL Server extension · Fabric database mirroring for Cosmos DB.
Graphs & ontologies — structure over similarity
- Ontology and graph databases: the missing link in enterprise AI and part 2: from theory to production reality — your most recent saves in the tag, and squarely on the Enterprise AI Architect beat.
- everything you need to know about graph databases (Neo4j) — the concepts primer.
- turning a relational database into a graph database with LLMs.
- knowledge graph vs. vector database — which to choose — the fork in the retrieval road.
- GQL is now an ISO standard — graph databases got their SQL moment, from the same committee.
- building a huge graph database at Netflix — 40 vertex types, 50 edge types · LIquid: LinkedIn’s large-scale relational graph database.
Production war stories
The section that keeps your DevOps instincts fed.
- Uber upgrades its MySQL fleet to 8.0 vs. GitHub does the same, differently — you saved them as a pair; read them as one.
- the Screener.in outage: a DELETE of 60M rows saturated disk I/O for 8+ hours — the cheapest database lesson is someone else’s.
- Pinterest’s new wide-column database on RocksDB — build-vs-buy at scale.
- TigerBeetle: the fastest and safest database — the talk that “set a HUGE standard.”
- Garnet — Microsoft’s Redis alternative, saved the week Redis changed its license.
- how the Spanner team does chaos testing · end-to-end tracing in Spanner — reliability engineering inside the database.
- Novo Nordisk: 12 weeks to 10 minutes with a MongoDB + LangChain RAG system — the enterprise case study.