DuckDB’s developers had more than a few ideas for improving how their database stores data, and some of those ideas went far enough that older versions might no longer be able to read files written by the new one.
That was enough to make October’s release DuckDB 2.0. Now the team is already working on 2.0.1. On The New Stack podcast, DuckLabs’ Gábor Szárnyas explains what made the changes worth a major release, what early users are teaching the team, and why he urges caution before moving a production workload to 2.0.
Codenamed Cinnamon Teal, DuckDB 2.0 is the analytical database’s biggest feature release to date, more than two years after 1.0.0.
We met with Szárnyas, head of developer relations at DuckLabs, at the company’s Amsterdam headquarters. DuckDB 2.0 can run as a server and query data faster.
“The other features are an extensible SQL parser, which was experimental earlier this year, but now is something that we turn on by default. That ideally doesn’t break anything, but it might,” Szárnyas says. “We also wanted to signify this with the release and the whole approach to doing more of a networked operation, shipping the Quack remote protocol as a first-class citizen in DuckDB.”
“Advice from database folks is that you should never migrate a .0 release in production. So I would like to repeat this warning,” he says. However, the Amsterdam HQ was all a-quack, with folks working to release the much more stable version 2.0.1 soon, “which will be a lot about bug fixes, small performance tweaks, and incorporating the feedback that we have received since 2.0.”
As Szárnyas put it: “In the eyes of many, all bugs are shallow.”
DuckDB 2.0.0 already includes dedicated Agent Mode, which makes it easier for AI coding agents to interpret errors and correct their own SQL queries, helping them get better results faster without wasting tokens.
“The AI doesn’t really care. The large language models don’t mind if a text is formatted badly or the punctuation isn’t perfect.”
“For AI agents, we just print simple Markdown tables in compact mode with no spacing. The AI doesn’t really care. The large language models don’t mind if a text is formatted badly or the punctuation isn’t perfect,” Szárnyas says, which lets DuckDB “compress these tables down to maybe 60, 70% of their original size. It is just as legible for the agent.”
Agent Mode comes with no hidden truncation, machine-readable errors structured in JSON, and guardrails against runaway queries, and it works with coding agents such as Claude Code, Codex, and Cursor.
Following Amazon’s announcement to acquire DuckLabs in August, AWS shared how it uses the in-process analytical engine internally, including to query in S3 Tables. Mai-Lan Tomsen Bukovec, AWS’s technology VP, says AWS has processed more than 2.5 billion queries using its custom Quick query engine with DuckDB integrations and optimizations. She writes that DuckDB has helped Amazon Quick reduce average query latency by 30%.
DuckDB is also set to become the universal SQL dialect of AWS Storage and Analytics. In its Q3 earnings call in October, Amazon CEO Andy Jassy remarked that this will simplify data processing with “a modern analytics engine that is fast, simple to use and ubiquitous, and where customers will be able to write SQL queries once and then execute using any AWS analytics service.” This includes zero migration or portability costs between query engines, which Jassy said will also help AI agents query data efficiently.
The post DuckDB just shipped its biggest release ever. Its own team says not to run it in production just yet. appeared first on The New Stack.

