--- title: "The Surprising Birth Of DuckDB ft. Co-creator Hannes Mühleisen | MotherDuck" description: "In this first Quack Chat episode, Mehdi interviews Hannes Mühleisen, co-creator of DuckDB, in Brussels during DuckCon." canonical: "https://motherduck.com/videos/the-surprising-birth-of-duckdb-ft-co-creator-hannes-muhleisen/" --- [BACK TO VIDEOS](/videos/) [YouTube](/videos/?category=YouTube#videos-and-webinars-library)[Interview](/videos/?category=Interview#videos-and-webinars-library)[Short](/videos/?category=Short#videos-and-webinars-library) # The Surprising Birth Of DuckDB ft. Co-creator Hannes Mühleisen 2023/03/06Featuring: [Mehdi Ouazza](/authors/mehdi-ouazza/) , [Hannes Mühleisen](/authors/hannes-muhleisen/) > **TL;DR:** DuckDB co-creator Hannes Mühleisen shares how the project started from conversations with the R community—who loved data but hated databases—leading to the invention of a new category: in-process analytical databases. ## The Origin Story At CWI (Dutch Research Institute for Computer Science), Hannes worked in the Database Architectures group researching how data systems should be built. A surprising discovery changed everything. ## "They Weren't Using Databases at All" When talking to the R community (data practitioners doing serious analytical work): - They stored data in CSV files - Some had hand-rolled data frame engines - Everything was slow and limited - **They really didn't like databases** Hannes was confused: "I really love databases. Why aren't they using our stuff?" ## The Two Problems ### Problem 1: Client Protocols Are Slow Traditional database client protocols (built in the '80s, never updated): - Row-based - Heavy serialization - Moving data between database and R shell was painfully slow **Fix**: They improved data transfer speeds. R users said: "Better, but..." ### Problem 2: Server Management is a Nightmare Client-server setup issues: - Works on your laptop, fails when sharing scripts - Configuration headaches - "We still hate managing database servers" **Insight**: What about SQLite's approach—an in-process library, no server—but for analytics? ## A New Category is Born > "It turns out we invented a whole new class of database systems: the in-process analytics thing." SQLite is amazing but designed for transactional workloads, not analytical ones. This category didn't exist. ## The Prototype: MonetDB Lite First attempt: Hack an existing system (MonetDB) into an in-process analytical database. **Result**: Worked well enough to get people excited, but hit architectural issues: - Can't just exit the process on errors (you're inside someone else's program) - Can't use APIs that change global state (working directory, locale) - "Really trivial stuff, but it adds up" ## Starting From Scratch Hannes and Mark Raasveldt made a bold decision: build a completely new database system. **The risk**: - "Monumental undertaking" - "Not successful 99% of the time" - "Not great for your career as a researcher—you disappear for 5 years not writing papers" **The advantage**: They already had product-market fit from the prototype. They knew what people wanted. ## The Grind > "Mark and I basically did nothing else—evenings, weekends—just hacking on DuckDB for a couple of years." SQL is complex. Supporting all the operators, edge cases, and details took massive effort. ## Open Source: 2019 Immediate positive feedback. The design resonated because it came from real user problems, not theoretical research. ## The Lesson > "My definition of success is not to write papers—it's to have impact. In data systems research, you have impact by making systems." The R community's willingness to try "crazy stuff that was barely working" and give honest feedback was essential. This kind of practitioner collaboration is rare in academia—but it's why DuckDB exists. TABLE OF CONTENTS - The Origin Story - "They Weren't Using Databases at All" - The Two Problems - A New Category is Born - The Prototype: MonetDB Lite - Starting From Scratch - The Grind - Open Source: 2019 - The Lesson Start using MotherDuck now! [Try 7 Days Free](https://auth.motherduck.com/authorize?app_source=web&response_type=code&client_id=bza3KWQpxRAFlTlRFXUo29AOg9xD7zcp&redirect_uri=https%3A%2F%2Fapp.motherduck.com%2F&state=STATE&auth_flow=signup&screen_hint=signup&ext-ph_distinct_id=ff58038a-0d2d-4ee1-80b9-d530431622bd) ## Related Videos [1:00:37](/videos/bi-midlife-crisis-ai/)[2026-07-28](/videos/bi-midlife-crisis-ai/) ### [BI Is Having a Midlife Crisis](/videos/bi-midlife-crisis-ai/) Jacob Matson and BI analyst Ryan Dolley on whether AI can replace your BI tool, what it actually solves, and what still needs governance. Interview BI & Visualization AI, ML and LLMs [54:22](/videos/duckdb-agent-moment-jordan-tigani/)[2026-06-18](/videos/duckdb-agent-moment-jordan-tigani/) ### [DuckDB's agent moment (Jordan Tigani)](/videos/duckdb-agent-moment-jordan-tigani/) Jordan Tigani helped build BigQuery, then left to bet that most data isn't actually big. Three years later, the agent era is making his case for him. In this conversation on The Analytics Engineering Podcast, he and host Tristan Handy talk about how MotherDuck and DuckDB fit together, why a single-node database makes more sense than you'd think for agents, and what it would actually look like to have a swarm of agents managing your data. YouTube AI, ML and LLMs Interview [28:00](/videos/duckdb-ai-modern-data-stack/)[2026-06-02](/videos/duckdb-ai-modern-data-stack/) ### [How MotherDuck Was Born: DuckDB, AI, and the Modern Data Stack | Jordan Tigani & Carly Kiser](/videos/duckdb-ai-modern-data-stack/) MotherDuck CEO Jordan Tigani on how AI is reshaping the modern data stack, why query latency now matters for agents, and the rise of open data formats. Interview AI, ML and LLMs MotherDuck Features Ecosystem [View all](/videos/)