Maximilian

The 2nd edition of my "React Key Concepts" book is available now"! Fully updated for React 19, including brand-new chapters on RSC, form & server actions, Suspense and much more! If you, like me, like picking up a book for learning from time to time and you're (like me), convinced that we'll still need (React) devs in a couple of years, this book will be perfect for you! PLUS: There are special bonus offers & packages available now for the launch!

Told you so yesterday: youtu.be/nLgmTis9uM0

The 4d chess move will be to find a CVE and wait with publishing it until you got a supply chain attack ready for the patches required by the CVE

Supply chain attacks are becoming more frequent - obviously also because of AI (easier to look for vulnerabilities, write malicious code AND more targets!). But imagine the fun you'll have once you got the first compromised AI agents going rogue on your systems... Not a question of IF ... just WHEN

No month (week?) without a new huge supply chain attack. We'll see more of those. AI obviously helps with finding attack vectors, creating malicious code, and escaping security mechanisms. But maybe more importantly, attacks are just so, so much more worth it. More software than ever is being written by more people than ever not knowing anything about what they're doing (or not caring). I'm pretty sure being an OSS maintainer has never been less fun. You're getting swamped by AI slop PRs and issues AND have to try to defend against malicious attacks like the one that's currently rolling over the npm and Python ecosystem.

@vanschneider This! You suddenly realize how horrible most coffee is

I made the mistake of learning how to make espresso at home. I want to be clear: I am not nerdy about it. I just like a clean 1:2 ratio. 18g in and 36g out at around 25 seconds extraction time. I use a WDT tool, but only because I respect the beans and don't want them to deal with any microscopic water channeling. I might even apply the RDT technique to my beans before grinding to eliminate static electricity. I use the force to tamp, but evenly and precisely at 25lbs because I don't want to seem like I'm crazy. The water I like to keep at 93 degrees for a medium roast, pressure at around 9 bars with an ideal pre-infusion time of about 3-4 seconds. Pre-heated cup I don't have to mention, because I am a normal, well-adjusted person who does not get nerdy about coffee. The only downside of all this is that most coffee shops are basically ruined for me now, for obvious reasons.

@badlogicgames GPT 5.5 loves those single line functions to death

feeling the AGI as GPT 5.5 on xhigh suggest to just introduce more single line functions that serve absolutely zero purpose other than looking kinda what was in the training set...

Going live on YT (maximilian-schwarzmueller) & Twitch (maxedapps) in a few minutes. Also almost at 80k subs. So amazing!

Think I hate on X: - "someone just did xyz" - "many such cases" - "not trying to be mean, just genuinely interested" Sorry, had to get that off my chest 😅

GitHub is facing a huge amount of problems. Uptime / reliability issues, serious bugs and now (yesterday) a critical security vulnerability that allowed remote code execution. Is it dying?

This is absolutely insane

@samehalyassin I hope not :D

@maxedapps is there new attack today?

Nowadays I get the feeling of playing Russian roulette every time I run "bun install". And yeah, I got a secure setup. Still - not super excited about getting that supply chain attack

@benvargas @badlogicgames Yeah, might switch. Love the self-improving aspect of pi. Got a bunch of „just prompt for it“ extensions running already. The extensibility is such a huge plus of pi

@maxedapps @badlogicgames you'll get more consistent results using the `before_provider_request` hook to properly set textVerbosity... it's worth the effort if this is a real issue for you. And as demonstrated by mario... it's not a lot of effort.

I thought a model couldn't get chattier than gpt 5.4 (using it in Pi). Well ... here we are with gpt 5.5...

@badlogicgames @benvargas Or simply AGENTS.md. Works well enough for me

@maxedapps @benvargas "pi, write me an extension that injects verbosity level x into the provider payload. make no mistakes"

@benvargas @badlogicgames Tried all modes. Changes thinking output but not the final output verbosity for me. Different prompts („be concise“) do. But out of the box I find gpt 5.5 (like 5.4) to be pretty verbose

@maxedapps Something that probably needs adjustment in Pi, if I recall, the Codex CLI defaults to `low` and being on `medium` would explain your experience. @badlogicgames

Since a couple of people suggested using a VPS instead of Vercel (after the Vercel security incident): Don't. There are many good reasons for using a VPS. I published a new "VPS Essentials" course this week. But security is not a strong reason for switching. Cost effectiveness / low cost (and predictability) and flexibility are better reasons. It's obviously a trade off. More details in the video.

@RaviTejaKNTS @MattReinholz Yes

@MattReinholz @maxedapps I guess he meant the model being verbose.

Called it in my 2026 outlook video btw => youtu.be/5WSS8Uzvy0k

@tushardave15 I see companies hiring more devs as they want to utilize AI-enhanced development. Vibe coding is not enough, you need those devs - for serious work at least

@maxedapps Where you see it heading in next 6 months

Shifting from big tech to everyone else?