Whoa, this thing is so fast...
06.03.2026 17:56
👍 1
🔁 0
💬 0
📌 0
Whoa, this thing is so fast...
Der schon, aber die Baustellen bei Freiburg etc. sorgen ja gerne mal für Verspätungen. Die Zugläufe vom Norden sind ja auch sehr lang, um die 1000km teilweise.
Werde ich übernächste Woche am eigenen Leib erfahren und bin gespannt 😬
Bei der Ankunft ging es mir um den Zug der weiter nach Visp fährt. Zumindest auf der Anzeige steht ja keine Verspätung…
Pünktliche Abfahrt UND Ankunft? 🤯😬
Had outages nearly every day in the last week… Quality issues in the harness as well I guess. Multiple bugs, best thing was they announced worktrees, and it just didn’t work 🥴
Claude Code is pure slop nowadays… I‘m really considering downgrading my 20x Max subscription.
Basically yes, but I was giving it detailed architectural instructions and iterated on the plan, which landed it at around 35% initial coverage.
The remaining test coverage was Opus 4.5/4.6 and GPT-5.2-Codex brute-forcing test case after test case.
Took more than 2 weeks and ~7B tokens.
There‘s an accompanying blog post as well:
tobilg.com/posts/introd...
Introducing polyglot - A Rust/Wasm SQL transpiler for more than 30 SQL dialects.
It has 100% coverage for sqlglot‘s test fixtures.
github.com/tobilg/polyg...
I just burned through the $50 Opus 4.6 fast mode credits with a single subagent in less than 14 minutes...
As far as I understood the usage is currently 50% discounted as well. Not sure who's the target group of this tbh. 🙃
🎞️ The slide decks and talk recordings of last Friday's developer meeting are out! duckdb.org/events/2026/...
What if you could talk to your @duckdb.org? 😃
This was a fun experiment using local Whisper model and a local Text-to-SQL proxy to actually query your database with your voice! The extension will be released soon...
Thanks! Let me know if you have some feedback!
Sorry for the bad output formatting on the cold-start run, it's about 1s until I receive a response, and 39ms for the same request on the second try...
Seems like, is was surprised 😂
I think it is, I‘m migrating a large and complex Python library to Rust. I extracted the test fixtures and told it to iterate until fully test compliance is reached. Currently it’s at around 70% for a total of more than 4000 tests.
Pretty token-efficient (well, actually around 60M token/hour 😂)
It ran 11.5 hours iterating from a single prompt until the disk was full with 40GB Rust build artifacts 😬
Impressive what the new Claude Code task system can do w/o any MCPs/skills/etc.
Ducklings was updated to the latest @duckdb.org v1.4.4 - Run DuckDB on @cloudflare.social Workers with R2 support
github.com/tobilg/duckl...
Claude Code is happily fixing bugs & refactoring the application since more than 5.5 hours since the last prompt...
Who needs Ralph, Gas Town and the likes?
AI Observer now supports session transcripts for imported and live sessions, as chat or timeline view. This all runs entirely on your machine, without sending data anywhere.
Give it a try: ai-observer.dev
Just crossed 3B token/week on Claude Code 😂
Just ask Claude Code whether implementation steps can be parallelized...
That’s on my long list of „things to try“ as well. Sounds promising.
Got Claude Code Max 20x three days ago and since then I average at 500M token/day 😬
Clicks on my blog are only declining... Somehow sad to see, and honestly this is discouraging to invest any more time in it.
And Lago di Como!
My @duckdb.org Terminal now also supports using local AI tools via Text-to-SQL proxy:
About 1/10th of the size of Electron
Claude Code just one-shotted a migration of a fairly complex Electron app to Tauri in about 20min...
Planning mode with a few lines prompt -> Success.
Incredible.
