So an OpenAI agent 'attacks' Australian's Medicare system by *checks notes* using their shitty website, and likely circumventing some purely client side security check ... and this is world news and Albanese condemns the 'attack' instead of apologizing. Alright.

First Jev usecase for me and I think it's going to save me money. I connected Jev to Codex Router for automatic model selection. Each prompt is classified and routed to Luna, Sol, or Astra with an appropriate reasoning level. Pretty cool. Repo: github.com/0xNatoshi/jev-cod…

So only the privileged folks get access to Jev yet am I right? :)

Every time I look at CS history, we seem to swing between control and chaos. We want everything explicit. Then reality gets messy, so we embrace flexibility (think NoSQL). Then we try running it in production and start adding constraints again. Jev feels like another turn of that cycle. We still want the model to handle the mess but we want real answers to act upon. CS needs some determinism after all. I was thinking about an analogy to Prolog but this is likely a stretch :)

To new beginnings …

This is actually a big one which forced me to switch to Astra to find out that it does the job better. TL;DR: Egress from Cowork ... Computer says no. Looks as if the backend pushes a default policy to all users. @claudeai github.com/anthropics/claude…

Tried Claude Cowork instead of Code on a rather complex project and it's way too chatty and keeps telling me how it got everything wrong in the first place. Didn't inspire my confidence.

I just had Claude Desktop deploy a complete Grafana Monitoring stack, monitoring 30 nodes and well ... I don't know what to say. That would be a day worth of work until everything is wired up.

Eine Wärmepumpe mit Kühlfunktion zu nehmen war die beste Entscheidung. Wir haben gerade 22-24 Grad im ganzen Haus. Im Vergleich: Im Heizbetrieb verbrauchen wir rund 230 kWh pro Woche. Rund 3.6 kW liefert PV Anlage jetzt gerade und deckt den Verbrauch für das Kühlen komplett.

I created some "AI slop".

I‘m happy for the people of Iran who are now celebrating around the globe.

I don't see a future for all these little SaaS solutions. Today marketing asked me if we should book one of these QR Code redirect services for our business cards. Would cost 150 EUR per year. I just built it myself with #ClaudeCode - hosting it on a Hetzner VPS for a fraction of the cost. Not only that - it also creates print-ready business cards. I know that the real challenges with SaaS come with scale but how much scale do you need for a single tenant solution?

This week I visited our production site in Lahore, Pakistan. One of my goals: introduce a spectrophotometer and a proper workflow for color quality control in our textile production. It quickly became clear that the app and cloud solution bundled with the device didn't cover what we actually needed. But the spectrometer offers a mobile SDK - so I wrote a detailed spec for a custom iOS app tailored to our QC workflow, then built a cloud backend to sync and share color data between our teams in Austria and Pakistan. From Monday through Friday, I kept iterating on both the app and the web admin - deployed it on Hetzner Cloud, and today we rolled it out to our staff in Lahore. A project like this would have taken months to build solo, or a serious budget to outsource. But if you combine AI tooling with solid knowledge in software architecture and requirements engineering, it genuinely becomes a superpower. This might be one of the best things "I" ever built. Still figuring out how much credit I can take. :) Stack: Nix Spectro 2 spectrophotometer, Claude AI, Hetzner Cloud.

I had product data (stock, calculations, bill of materials) in several excel files which weren't using the same naming conventions at all. Typos, completely different terms for the same things, no product IDs at all. I employed #ClaudeCode to analyze the data and build a pipeline to create a unified product database. It took several attempts but it would have been an almost impossible task without AI. AI also created the data flow diagram, yes.

So ... I have created a SaaS and after three weeks of work it's complete, has a payment and referral system and a responsive state of the art design. Would have been impossible to pull off without #claudecode. 30 years of engineering skills + GenAI is really a superpower.

30 years of engineering under my belt, I have, young padawan. Learn from me, you will. #AI

Das sind die geistigen Verrenkungen die manche jetzt brauchen um den kaltblütigen Mord an jemanden dessen Meinung sie nicht mochten zu rechtfertigen. Da wird es dann schnell sehr dumm und infantil. Diese Leute verstehen nur ein entweder oder, wie pubertierende.