Martin Mazur

Solving problems with LLM.

10h in to modifying prompt for problem that would take 2h to solve by hand.

I remember when setting up computers was fun. Is this a sign of old age?

Just got myself a M4 MacBook Air. Nice upgrade from my M1 but setting everything up is a pain.

I don’t want to copy all the junk of my old computer.

Build progress so far.

3 sets completed. I’m amazed that most of the pieces are still there after years of creative building.

Kids wanted to re-build some of the old Lego sets.

Up until 2 am sorting and finding pieces.

Rain day today - we are building at full capacity!

We still want feature films even if we consume YouTube and TikTok. It’s the same content type but serves different purposes.

Home built and enthusiasts software will be different from big company software.

The tricky part is to figure out what hat will be pushed away and what will stay.

Big movie and music studios employ slightly more people now than before YouTube and Spotify democratized their industry.

We have more content but the old jobs are still there.

If the new tools democratize software development. We will have more creators - but the old jobs will still be there.

This is pretty fantastic.

The timer in question is at least 10 years old and using search would surely have led me to several questionable sites with scanned pdfs and a lot of tears and frustration.

(Yes, the instructions worked)

Eliminating the invisible.

A big part of product work is eliminating risk. Proving the value of this kind of work can sometimes be tricky. It does not amass to a body of work or a deliverable - instead it makes everyone else succeeded.

Honestly, my use case right now is to ”hate css less”

It’s a very slim use case.

But I’ll keep everyone updated.

Using a Coding Assistant is the new Resharper. Back when Resharper was new, lots of people refused to use it, saying it suggested the wrong things, that it made people lazy. Some of that was (and is) true, but on the whole, it's not.

If you know how to code, assistants can accelerate a solo dev.

I have Gemini via GWorkspace as well so should try that.

I’m on a MBP M1 so anything beyond smaller models becomes slow.

Besides, I don’t have the time to tinker - for this I want it to ”just work”

Will have to try Claude as well. Haven’t even considered that.

Haven’t even considered that. Will give it a spin! Thanks.

Self hosting in - running a smaller model locally or? If tried that but I don’t find it worth the hassle.

How do you use Gemini? Integrated with an editor or just ask it questions stand-alone?

Cursors or VSCode w/ co-pilot?

I code that much anymore but it happens once in a while.

I’m considering getting a paid plan for a AI coding tool.

Which one do you like and why?
Is cursor worth the extra $100/yearly over vscode?

”Everything is going well and then it randomly throws in the word Globe. WTF?!”

OH two devs trying to debug their AI code.

We are living with uncertainty not only in the world, but also in the systems we build. With LLM output can be anything.

Rigor and curiosity Kids grow up with innate curiosity. It’s the hardwired instinct that permits us to walk, talk and survive long before we get to school. And at school, the industrial imperative prizes rigor o…

We were all curious at one point in time - what would it take to rekindle that curiosity in our organizations?

seths.blog/2025/06/rigo...

💡The Apple Hint Everyone Missed Apple is training us for a post-iPhone world

Is Apple training us for a post smartphone world?

@sineadbovell.bsky.social on the hint everyone missed at WWDC - here is what Liquid Glass is really about!

open.substack.com/pub/sineadbo...

The AI crowd screams loudly : "some kid using AI will replace you!"

I am not afraid of that kid, I am afraid for that kid.

I was doing my finances and mid everything Gemini started to say that it’s not allowed to give me bookkeeping guidance since it’s only a language model.

LLMs are weird. Don’t take it too much to heart.

Picked up running again, during 2 week break from Muay Thai.

I haven’t done running since 2023. My second run this year - 6k

Best use case for LLMs is writing documents that are required for full filling a process and that, I know, nobody will read anyway.

This has previously been tedious work that now is extremely smooth.

Programmers don’t want you to know about this one trick to fix all your bugs in one go

✨ We're making NewCrafts Conference Talks accessible to everyone. Explore our archive of thought-provoking talks on software craftsmanship, now on YouTube. Subscribe to stay updated: buff.ly/SNm4XES

😂 - maybe, or I just choose not to remember.

Maybe not.

But I would think the tutorial would scp the cert to the server.

It’s just as easy (if not easier) and it’s more secure.

Also, the tutorial would not give you a bad practice and then later tell you not to use it.

Oh - we are in for a world of pain with all the new AI assisted developers.

ChatGPT: Also ChatGPT:

Nice! Also found this to complete the setup.

kill-the-newsletter.com

I’m getting close to something ”feature complete”

I’m considering this setup but using a paid miniflux (miniflux.app) instance. $15/year to save me a lot of pain.

I tried Feedly but miss some flexibility in it.

Are you on the free tier?