Mathias Lafeldt

Indie product engineer obsessed with dev tools, performance & AI.

Maintainer of duckdb-rs & duckdb-go.

Zig-pilled. ⚡️

[Mathias Lafeldt @mlafeldt]

One of the benefits of maintaining two DuckDB client libs is that I keep finding and fixing bugs in DuckDB itself

https://t.co/KSG5AeA2Hm

[Mathias Lafeldt @mlafeldt]

@mattpocockuk This skill does it already, using copy instead of symlink in postinstall

https://t.co/DIgxCvILuf

[Mathias Lafeldt @mlafeldt]

PR: https://t.co/nxZnHO44GL

[Mathias Lafeldt @mlafeldt]

Good morning!

Starting the day by deleting 1500 LOC from duckdb-rs 💪

[Mathias Lafeldt @mlafeldt]

This one: https://t.co/NG9Dg4LQeu

[Mathias Lafeldt @mlafeldt]

Using /grill-me to figure out how to classify my freelance work for the client's "new invoicing system"... 😑

[Matt Pocock @mattpocockuk]

Writing ADR's for agents has been such a good decision

Capturing all the non-obvious decisions in a codebase makes every agent in your stack that touches the stack smarter

It's the thinnest layer of docs that captures the stuff code can't

[Mathias Lafeldt @mlafeldt]

My example uses the thermo nuclear review skill, which is really good but also really token hungry. Watch this video from @mattpocockuk to learn more.

https://t.co/p1RPLbXnil

[Mathias Lafeldt @mlafeldt]

Took dynamic workflows for a spin. Took me a while to get the prompt right: disable-model-invocation prevents spawned agents from invoking skills via slash commands (which is confusing/wrong). Happy with the result once all is working, though workflows are a real token burner!

[Matt Pocock @mattpocockuk]

Cursor shipped a /thermo-nuclear-code-review for the TOUGHEST AI code review possible.

But is it any good? Let's dive in:

[Armin Ronacher ⇌ @mitsuhiko]

This is such a good post. https://t.co/IdmAnh18Nt

[Mathias Lafeldt @mlafeldt]

@mattpocockuk 👏

[Mathias Lafeldt @mlafeldt]

The Codex App has removed the context usage indicator from its UI. Apparently because auto-compaction is supposed to just work. Bad move in my opinion.

https://t.co/Q2Y2OwG6Xm

Context pollution is still a thing: https://t.co/PJYXlrcH55 by @mattpocockuk

[Mathias Lafeldt @mlafeldt]

@mattpocockuk "They removed the thing that helped users deal with the current reality because it didn't match the future they wanted to project....but before actually building that future!!"

Well put.

[Mathias Lafeldt @mlafeldt]

Claude starting the Docker daemon for debugging without asking me is both unsettling and cool (exactly what I would have done anyway).

[Mathias Lafeldt @mlafeldt]

In contrast, Claude Code's new /code-review is pretty good, even though there's a bug that causes findings to be reported in JSON.

[Mathias Lafeldt @mlafeldt]

Does the @coderabbitai CLI actually find anything useful on the free plan? Always reports "No findings" for me.

[Mathias Lafeldt @mlafeldt]

@HadesArchitect @coderabbitai Yep (I know how git works)

[Mathias Lafeldt @mlafeldt]

Does the @coderabbitai CLI actually find anything useful on the free plan? Always reports "No findings" for me.

[Mathias Lafeldt @mlafeldt]

I use /improve-codebase-architecture and /prototype almost every day, and have created my own skills based on patterns from this repo (ADRs are a real game changer). Highly recommend. The 100k+ stars are absolutely well-deserved.

https://t.co/JJ8UV7eRrr

[Siddhartha Saxena @siddsax]

Anthropic onboarding day: Michael Scott introducing Karpathy like he just signed Wemby in free agency.

[Mathias Lafeldt @mlafeldt]

@duckdb Just noticed that my DISABLE_EXTENSION_LOAD fix made it into the release post.

https://t.co/OOVuI4QcHr

[Mathias Lafeldt @mlafeldt]

I've contributed 5 PRs to this @duckdb release. Only one of them was reverted almost immediately. 😄🤙

https://t.co/EVxUBafzYM

[Mathias Lafeldt @mlafeldt]

scc --avg-wage 150000 reports that 13k lines of Rust code amount to an "Estimated Cost to Develop" of $1M. Not unreasonable.

However, 1500 lines of (AI-generated) Markdown will add another $200k.

Pre-LLM cost estimates are (even more) fiction.