I wonder if we will see a rise of RUP (Rational Unified Process) and similar heavy-weight docs focused methodologies to better instruct AI agents and when turnaround time will be minimal

Being able to put a chain of AI-agents to work on creating PoCs in like 30 minutes with full test coverage and suprisingly good code quality is wild. But - it's like creating a conference talk. 1 minute of talk = 1 hour of prep. So 30 minutes of AI work = 30 hours of human finish. Not

Idk what it is but I own two Hans Zimmer vinyl records. One is the Interstellar soundtrack and the other is the Dune Part 2 soundtrack. Love Zimmer but both records came with factory defects. Not ok since I had to sell one of my children to pay for the records.

The irony that Microsoft blocks access to AI agents fetching webpage data when pushing agents/CoPilot as the future of everything

One thing I've noticed when working with AI agents is how extremely easy it is to get sidetracked with scopecreep when you can just fire off another agent to go do something. The age old "focus on one thing at a time" is still relevant as agents never actually 100% finish the tasks for me.

Immich is such a promising project. After running it for a few versions I have concluded that it's not yet read for primetime production use (for me at least).

Omarchy 3, now with a red dot.

How have I not owned a 3d printer before now? My cable management game is about to hit a new level.

Somehow gen ai lead to it being acceptable to say "Yeah I'm not sure the content in the document is totally correct". Not a good thing. You deliver it you own it. Doesn't matter how you created it.

More and more "normal people" I talk to keep saying that "I'm starting to doubt this whole AI hype. We are not really seeing any tangible results".

Having a proxmox host going 0 byte disk sure was interesting. So much strange things happening when disk runs out - ended up getting to bored and just rebooted after putting disk back.

It's gonna be a long time before letting LLMs run normal people browsers will be a non idiotic thing to do. "Our testing has identified scenarios where Claude could be manipulated to: Extract and share sensitive information with bad actors. Delete important files. Perform unintended

Tip of the day: Remember to set the right power profile when working on a laptop. "Balanced" is turtleemoji throttling

Just found out about floating windows in Zellij. So nice for a (or more!) bash scratchpads that can be toggled on/off with a keybind. Great stuff.

A hosts-file with all random AI domains would be helpful. Should start blocking at DNS level..

Doing kernel compiles on 4% battery. Is this peak #thuglife as ... man that remembers his youth?

Also pressing jl<space>w to go to the browser somehow just also feels like a major productivity win. Even if you count the ROI of setting up and learning to use the keybinds + the loss of typing speed - the loss of general productivity... I mean what negative ROI can compensate for that

I don't know what it is but something about mistyping commands into a terminal makes me feel in control of my enviornment.

Next halloween I will go as a "vibe sysadminner". Now thats scary.

Anthropic just changed their privacy settings to "train on my data" now being opt-out it seems. Both for web and for claude code. So much for being privacy focused.

all these browser agent are going to generate so much chaos and crap... poor early adopters who just thought it was cool and didn't think about all the ai exfiltration we have been seeing lately. https://techcrunch.com/2025/08/26/anthropic-launches-a-claude-ai-agent-that-lives-in-chrome

Every time I use gpt-5 I get so underwhelmed and go back to ... any other model.

It's interesting the way cli coding tools read files like 'Read src/services/content_manager.py (limit=20, offset=95)' might mean that splitting things into short, readable chunks is primarly for the benefit of humans, AI doesn't seem to care.

1password - still figthing me at every turn. Why. I just want to use you.

I use a tiny CLI called bs to get thoughts out fast. It reads from args, stdin, or an editor, then posts to Bluesky and Mastodon. No ceremony, just words out.

bs updated: removed like mode and URL parsing. Posting only (Bluesky + Mastodon). Simplified CLI and code structure. This is a test of bs posting.

The recent spam is from my lovely 'bs' program that comes as part of my omarchy-pbjorklund project. https://github.com/pbjorklund/omarchy-pbjorklund/blob/master/bin/bs

New zen-browser update added folders and having used them for 5 minutes I realize I have missed them for as long as I have been using this wonderful piece of software.

Testing the fixed threading implementation! This message should be long enough to split into two parts and create a proper thread on Bluesky. The first part posts normally, and the second part should reply to the first post using the correct ReplyRef structure with parent and root

My god getting kanata up with a second layer letting me user altgr+['; to get åäö actually had me learn som lisp.

AI investments give terrible ROI so far. Read: goldmansachs.com/insights/top-of-mind/gen-ai-too-much-spend-too-little-benefit mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai goldmansachs.com/insights/top-of-mind/gen-ai-too-much-spend-too-little-benefit

Cool, got my own mastodon server up. Not sure why, but choosing a server gave me to much analysis paralysis.

The key skill in vibe coding is noticing when the LLM starts struggling and starts looping or just going the wrong way or where it obviously didn't have stellar training data.

Ok revanced for android is pretty cool.

Ok zellij is kinda nice. Tmux out, zellij in.

My dotfiles are now complete and perfect - said noone ever that has successfully quit vim at least once.

If I could just get that TUI browser that can interpret JS I will have sucessfully emacsed my nvim setup

To go with a new standard mechanical #keyboard or a ortolinear split one... Thats the question of the day.

Looking at pictures of old programmers near their vac tubes and punchcards. Is this how the youngones are looking at current day programmers?

Now all I need is a terminal browser that can execute javascript correctly

I love my pbp check command

Got to love programming back in the day because I could tell the computer exactly what to do and it would do it. Now I tell LLMs what to do and it tells me I'm absolutely right! And then proceeds to not do exactly what I told it. They call this progress.

Remember kids, don't delete your terraform state. It's a world of hurt.

It's kinda amazing I can use my swedish digital ID 1 second from behind a VPN in Netherlands and 2 seconds later from Sweden without anyone complaining.

This whole "let's talk it through" thing feels inefficient

Speculation: AI companies are pushing the type, showing incremental improvements at the application layer while feverishly funding their R&D departments to come up with the "next transformer architecture" that actually CAN live up to the hype.

Cool it works. Could probably make it more interesting by running the suggestions through some LLM as well. Current scoring (0-100): Mutuals: 40pts max (8 per mutual) Quality: 25pts (profile completeness, ratios) Categories: 20pts (bio keywords) Influence: 15pts (follower count 0-10k scale)

Now that I found some people to follow on bluesky I'm working on my script to build a graph of those followers so I can see who else to follow. Ah, computers.

Back in the day people got in big trouble for downloading a single song from Napster. Now we have LLMs built on blatantly stealing all the worlds intellectual property without anyone seeming to care. So piracy OK now? Or do I need a few billion in VC cash first?

Followed a huge number of accounts, just to have something. I'm not a bot. Promise.