SAM-3D looks awesome. Hope I find some time in the next weeks to play around with it creating 3D reconstructions of the world around me. Looks like it should run locally as well. ai.meta.com/blog/sam-3d/

Oh boy, do I regret updating my Macs to Sequoia. Seriously planning to downgrade, even though it means wiping the machines completely. While the update is bad, the other half of the truth is that I tried out so many tools in the last year, it's a mess. www.macworld.com/article/6713...

Here's an impression of our last meetup on Tuesday!

And another full house at our Agentic Coding Meetup #3! It was good to meet so many new and familiar faces at the Meetup, all coming from diverse backgrounds! We also have a new website for the meetup now! agentic.hamburg

Why you should buy an AMD machine for local LLM inference in 2025.This is the hardware I chose for my home LLM inference server. Here's why. www.stefanmunz.com/posts/2025/w...

One hour into testing Google Antigravity, I reached the quota limit. Interestingly no upsell option offered. Else it's a really good experience, I especially like the UX of the I like the hands-off Agent Manager. No opinion about the quality of the code yet.

Only one week to go, then finally we meet again! The next chapter of Hamburgs Agentic Coding Meetup is just around the corner. We have 4 seats left on Luma luma.com/dicrp3xl Also a few left on Meetup, grab them while they ara available: www.meetup.com/agentic-codi...

Open Source and self-hostable LLM tracing and evaluation platform, built on the OpenTelemetry standard. phoenix.arize.com

The Chicago Federal Reserve Bank on why the main productivity gains in the US economy are due to the IT sector and why this concentration bears risks if it continues like this. Seems to be rather unprecedented what we're experiencing right now economically. www.chicagofed.org/publications...

What a strategy: Building a private equity company, applying their playbook and using VC money to finance it. The finance and startup world continues to amaze me, but in a negative way. As a customer of meetup.com, a company Bending Spoons acquired, I'm not confident this is a good playbook.

Why you shouldn't buy into the Apple ecosystem for local LLM inference. Apple Silicon Macs are engineering marvels. But here's the hard truth: Apple's RAM pricing model makes their hardware prohibitively expensive for local LLM inference. Read the full article: www.stefanmunz.com/posts/2025/w...

HUMANS WERE NOT SUPPOSED TO "SYNERGIZE" OR "GIVE FEEDBACK" Love this controversial advice, especially right now where it's about experimenting with AI instead of strategizing or stakeholder management. newsletter.posthog.com/p/collaborat...

omg ist das geil! 🔥 Digitec: " Warum wir unseren Devs mehr vertrauen als Big Tech" #leadingByExample www.digitec.ch/en/page/digi...

I love the transparent and direct communication style of the rePebble project. Really looking forward to receive my model early next year! ericmigi.com/blog/how-to-...

Köln, I am coming! Looking forward to meet a lot of new people at the Digitale Leute Summit tomorrow. Super happy to be part of their trainer team for their upcoming AI Software Engineering Bootcamp mext year! www.digitale-leute.de/summit/25/

Great to see Charity Majors live on stage finally, I‘m a huge fan! In praise of normal engineers, the perfect topic at this time!

Agentic Coding Meetup Hamburg #3, we're coming! Extremely proud to host 4 speakers this time, 3 of them first time speakers at our format! luma.com/dicrp3xl

This sounds like easy advice, but I find it really hard to follow. It's just so easy to allow new facts to change your thinking. Sometimes it's good and sometimes it's bad. designobserver.com/the-compound...

A very different perspective on the same AI world that I am investigating. Interesting to get more insights into the financial aspects of it all. www.stateof.ai/2025-report-...

Interesting walkthrough on how to implement memory for LLMs. I am still searching for a good intro how these memories work and why. www.philschmid.de/gemini-with-...

“Regulation Isn't the European Trap — Resignation Is […] When you're step 3 of 10, act like step 10 depends on you. Own the handoff. Move same day.” lucumr.pocoo.org/2025/10/21/e...

I switched from the 180 Euro subscription of Claude Code to the one for 90 Euro. Claude Code is still great, but I'd like to force myself to try the other agents more this month.

Wow, I thought Claude Code supported AGENTS.md instead of CLAUDE.md . Instead the simple to implement change is still open... github.com/anthropics/c...

TIL about asdf, the version manager for multiple programming languages and plugins. Written in Go, looks great, will try it out immediately. What finally convinced me to give it a go is the ballad of asdf: github.com/asdf-vm/asdf...

Apple is led by finance people. Good if you're a short to mid-term investor (as the finance execs mostly are), bad for users of the products. Time to start looking for alternatives. www.youtube.com/watch?v=JUG1...

Stacked diffs have existed for a while but never became popular. With agentic coding, they make more sense, because it's easy to change many things in one session, which becomes hard to review. www.youtube.com/watch?v=qBRJ...

A job board where you own the data is a fascinating idea. I hope this takes off. atwork.place

Dug into the self-hosting part of the AT protocol. It's on my list of things to set up, even just to understand it better. atproto.com/guides/self-...

"The need for a patched mesa comes from the fact that virtio-gpu doesn't have the ability to negotiate blob alignment with the host" I mean, what does this even mean? I love virtualization with GPU access, not! github.com/apple/contai...

Cutest moment with Claude Code this week: Trying to find out how much power this home server draws, only to find out Claude is the top consumer right now. "That's me!"

💪 Finally, it works. This took longer than expected. The infrastructure for self updates with Go is awesome. It is still a hassle to implement, unfortunately. Now it works, two production systems up and running. Fingers crossed no agent breaks this code tomorrow...

I love how new terms are being coined at the moment. Simon Willison's post about getting from Vibe coding to Vibe engineering is a perfect example. Unfortunately, it's missing one key property of codebases: The differing viscosities of your files. Read the full article with links in the link!

Current Status: Comparing 4 reviews from LLMs for my newest newsletter article, coming out tomorrow. I keep coming back to Gemini 2.5. Pro for these tasks. Claude and ChatGPT answers often feel rushed, as if they didn't really take the time to read the piece. Sloppy somehow. New in the mix is ...

Beautiful article on the web's origins and the founder's vision. I totally agree we can restore the web to its intended state compared to this monopolistic shithole we are in right now. www.theguardian.com/technology/2...

Great advice from the PostHog team on avoiding AI coding mistakes. We'll need to structure code bases very differently use coding agents effectively. Also, scripting languages like Python are much harder to improve with agents. Happy I can build a code base from scratch for agentic needs...

Switched to the glass releases of iOS and macOS this weekend. I dislike the different corner radiuses (radii?) per app on macOS, that just feels wrong. Else it's ok, but not a refresh that I needed. Disappointed and sorry for all the iOS devs that needed to adapt their codebases to this.

Exploring Stacked Diffs with Graphite this week. With the power of coding agents its incredibly easy to change a lot of things in one session, but it's getting way too hard to review. Let's see if stacked diffs are an answer to that. graphite.dev

This has the brutal undertone you would expect of a 996 company. It leaves me with the impression they're working against and not with the model. cognition.ai/blog/devin-s...

Interesting to see how the system prompt changed in Claude Code, but only for Sonnet 4.5. Apparently, they are able to fine-tune the models much more to their needs. Let's see if other vendors follow. Ultimately, this could mean that models and coding agent software become much more interchangeable

What an insightful piece. I think Armin took a valid approach; I'm more and more seeing the negative effects of delegating 100% of code to the agents. lucumr.pocoo.org/2025/9/29/90...

A few weeks ago, I showed you custom slash commands for storing prompts you need repeatedly. But I ran into a problem: sometimes Claude followed my instructions perfectly, sometimes not. I found the fix. Read the full article with links here: www.stefanmunz.com/posts/2025/c...

Claude realizing by itself that this is getting too messy. I'd like to see more of that!

Phew, who would have thought that building webapps with simple frameworks like Astro also gets really complicated quickly. Today on the menu of things I did not know about and did not plan to know about: Content-Security-Policy and where to handle it...

vLLM support is coming for the AMD AI 395 Ryzen processor, most prominently the Framework Desktop computers. Awesome, can't wait! github.com/vllm-project...

I sometimes watch these hardware videos, just to remind myself how painful hardware is compared to software. The main message is a bit hidden: It rarely makes sense to cluster computers for a task. Each node must be able to process requests individually, leading to parallelization. This was true...

"The main bottleneck for building software (and other systems) is the time required to let the ideas bounce around in our head until we feel that we have a good grasp of what direction to take." Absolutely nailed it, I agree! openshovelshack.com/blog/let-it-...

The death of the corporate job is an evergreen topic, of course. But so far, people lacked to see a realistic alternative to their corporate pointlessness. It really has become easier to run a one-man-show with AI. Let's see where this goes! thestillwandering.substack.com/p/the-death-...

Finally my Framework Desktop machine has shipped, can't wait for it! frame.work/de/en/desktop

Game Over for the week, at least with Codex. Surprisingly good value for 20 dollars per month, highly recommended! Not as addictive as Claude Code, but in a good way.

No, the internet is not decentralized. It was maybe a long time ago, but shut down the servers of the Mag 7 and the internet for the world as we know it is gone. He is right though, it should be more decentral again, and I'm already working on that. daringfireball.net/linked/2025/...