Marc Klingen

⭐️ 10,000 Stars! 🤩 We've just crossed 10,000 stars on GitHub! A huge thank you to each and every one of you for your support. (thread what we already shipped in 2025 👇)

So excited about voice-native interfaces Everything is better voice-first: email, support, writing docs/specs, providing lots of context to cursor

getting there please steal the ci checks for your own docs, nothing's worse than links not working (external ones pointing to your docs, or internal references) check-*.js here: github.com/langfuse/langf…

Hoping that I won’t screw our SEO by making this huge docs change Added way too many vibe-coded tests that check links (internal, compared to prod Sitemap, redirects), hoping that I’m safe Excited for this to land

Will publish our little “agent eval 101” based on what we learned from working with the @langfuse community this weekend

Recently overheard In the past, enterprises invested years into breaking down their data/software silos If a human can log into multiple silos to complete a job, AI can do it as well Thus AI projects aren’t blocked by silos (start now) + reduced ROI of breaking down silos

a good metric for ai adoption in a company is how many non technical people are pushing code to production

"failure mode taxonomy" is a good abstraction

It was awesome to get such a fast and exciting onboarding to langfuse💪

100%. Next collision-installation done Lots of ideas to improve onboarding and docs Thx team @stagewise_io, this was fun

If you have questions, ask Matt and you might get a really nice video answering them

if you do "forward deployed eng", there won't be the sensation of waking up to new ARR and happy customers

everyone talks about "forward deployed engineering" seems like a trap as it has a very clear path to revenue when all enterprises want "agents" huge risk of doing lots of custom work that cannot be productized love forward deploying myself though for product research

8h sleep (today): love support, writing docs, keyboard shortcuts are super fast, all meetings are very productive/creative, <6h sleep: fml Will try to remember this, no chance that being able to work 2h more (input) is worth it

Sharing this essay way too often recently > Make good new things Helps escape the midwit trap, so much opportunity right now, so little filter necessary to build something that works positive side effect: optimizes for personal interest and energy

Looking at traces is all you need If you have way too many, add some user feedback metrics or online evals to flag the exciting errors in prod

Forgot how much I love whiteboarding with customers, this week is already super interesting/fun