Building software since at least yesterday, 4 products, all MRR. Strong opinions, loves a joke. Netherlands 🇳🇱

Netherlands
Replying to @claudeai
@claudeai Opus 5.5 did something very cool. I have a charging station at home that would have been migrated to a new backend provider. That failed and it was out of order for weeks. After long time calls, the solution was, "Sorry we can't help you, good luck."
1
1
40
We got the admin password to the station, so with an App we would be able to reconfigure it. But that app didn't work, due to (probably) the same reason as why. the migration failed. So I spun up a Claude Code session, pointed #Opus55 to it and i let it run.
1
13
In about a session of one hour (I was also doing other things simultaneously), it figured it out. It fixed the connection issue (expired TLS cert), found the firmware bug, rewrote the configuration, and made it work again. Incredible.
6
What I did do this weekend: I revamped a PC I had lying around to a Linux server, that is now my remote dev box... Full YOLO mode, no risk of deleting my laptop, 5 features being built at the same time... I'm ramping up speed. All Opus 5.5s are nowhere near the limits.
107
LOL.. Claude: There are still some blockers left; besides that, we are ready to merge. Me: What blockers?? WDYM?! Claude: "Blockers" is the reviewers' word for problems that must be fixed before this PR can be merged. They are security holes or things that break in production.
187
I see many people being concerned about worktrees. Being too resource-heavy. My setup: 5 permanent worktree's on disk. All with their own local dev environment. Agents just switch a branch within that worktree, plus a small log to keep track. Never overhead again.
1
135
This morning I woke up to a @posthog billing alert. "You are due $400 this month." Never got out of bed that quick. How can that be? So i dived in, and it is caused by 1 customer, running the cli of @envmanager in a loop, causing 1.5 million events...
2
7
3,789
Quickly updating the cli, adding a block on the server to hold it back, and let's hope there not more of these funny perks in my code...
273
Kinda quiet from the "open" side right now...
156
Where is that Codex reset....
1
228
What a difference between then... and now.. #chatgtp
1
41
Anyone else noticing, Claude (Opus especially) starts responding in your native tongue, without any instruction to do so?
29
Patrick Gerrits retweeted
Replying to @itsloganbutler
LOL! Can you train my kids to cleanup their room every now and then?
1
1
2
43
So I see numerous people talk about AI slop text. Over the last few months i've created a skill that i 'trained' on my writing style, my texts, and how i like to communicate. None of the AI detectors detect it (100% human) and my customers dont seem to notice.
1
3
74
Nobody cares, I think, but with a good skill, you can get very very far. It is very easy to create this. Just ask your model to create a skill for your copy-voice. And ask it to generate something. Then you correct/redact and give it back and ask it to modify the skill based on your corrections.
1
1
56
Then keep doing that for a few weeks/months and you have build a skill, with a decision log, that the ai uses to write content. Works really well for me.
6
I really wonder how the algorithm picked me for this email… i do like tempura shrimp…
29
I really can't believe that people actually think Fable is GOOD at design. It's decent at best. It just designed an alert with a button inside it to take action on the alert. You know what that button does? Open a second alert below it.
1
29
It is so funny that on tech X people shit on Copilot all the time, but at the same time it is widely used at many enterprises. Just because you, sitting in the basement, do not use it doesn't mean nobody does.
27
This whole pacing the frontier is so funny. By spinning this, letting us believe they should pace, it enforces the idea in our heads that they are miles ahead, have AGI, and are going to wipe out the earth. Without us knowing anything about their future capabilities.
2
1
51
Maybe they hit a plateau, and there is no next better model. Perhaps they already have a sentient AI model running in their lab. We don't know. So all i can think about is this huge marketing campaign called "pace the frontier".
1
31