Asking my personal AI assistant to install a plugin which enables it to read my Google Calendar is just amazing. Seriously, give it a try: github.com/skorokithaki...

1/ I've been working on Stavrobot (github.com/skorokithaki...) fervently over the past few days with a custom OpenCode architect/developer/reviewer/reviewer harness, and I am amazed at how fantastically few bugs I've seen.

Of course I had to make my own personal assistant agent: github.com/skorokithaki... It's more secure than OpenClaw, but perhaps not as competent yet. I really like it, though.

I couldn't wait to get my Pebble Index and made a pendant with a button that I can use to record my thoughts and notes. I also made an Android app that will receive, store, and transcribe those notes. I'm really liking this, I'll write it up soon.

I really hate how I have to log in EVERY SINGLE DAY to the 100 sites I use, because someone opined that "long-running sessions are bad for security". I ended up disabling 2FA everywhere because I couldn't deal with wasting 10 min a day logging in. @owasp.org please update your guidance.

Claude today did something I've never seen before: it basically went "yeah man no idea" and asked if I want to talk about something it *did* know about.

After months of LLM-only programming, I wrote a program by hand again today. It was nice for a change, I think I'm going to do it again.

Is GPT-5.2 Codex *way* too snarky for anyone else? It's been giving me lip for ages, and now it made a mistake and I got fed up at the snark and demanded it apologise. This is its refusal.

The code in all my new projects is 200% written by an LLM. The trick is knowing which 100% to keep.

Before I spend a bunch of time reinventing this wheel for the hundredth time, does anyone know of a simple 4x4, nice!nano/ProMicro-based macropad PCB? If not, do you want one to exist?

I made another little bedside clock, and I love it: www.stavros.io/posts/i-made...

If you're interested in making things/3D printing/circuits/software/whatever, our making community has a low-volume newsletter you can sign up for: www.themakery.cc

@profanity.accountant do me!

I don't understand how Google can be smart enough to create the world's most powerful LLM, but not smart enough to make an LLM web app that doesn't lose your message when you navigate away for a second. Seriously, what's going on?

I love how we live in an era where we need to defend against social engineering attacks on *our programs*.

I had a spare rotary phone and some meetings: www.stavros.io/posts/i-conv...

I polished and published a #MicroPython experiment for emulating OLED displays in the terminal. Uses kitty's Terminal Graphics Protocol. Feedback welcome!

I had some spare time so I made an encyclopedia: encyclopedai.stavros.io

I made a site that vibe-codes itself: theboard.stavros.io

I am making a tiny LED panel and it's going to be amazing.

I'm making a new thing, does anyone want to give me feedback? pine.town/new

Fun fact: I once crashed reddit with an April Fool's joke and then I forgot about it.

I wrote (yet another) Jujutsu tutorial: www.stavros.io/posts/switch... I haven't published it yet, I'm looking for feedback first. Please let me know if it helps at all, or if you're still having trouble with JJ's mental model. I'm worried I've written yet another unhelpful thing.

This is my regularly-scheduled PSA to gush about how much joy Jujutsu gives me, something I could never say about git. It's truly fantastic, and jjui just completes the whole ensemble. Fantastic pieces of software, switch to jj today if you haven't yet!

Wow, the guy from Valetudo banned me from their Telegram channel for a year because I dared to ask about some behaviour of my vacuum, running Valetudo. I literally sent one message, they debated a bit about whether it was relevant, and then banned me for a year. Nice.

I wanted to look at my blood test results in a way that answered my questions, so I made a small thing to ingest my spreadsheet of them and show them in a nice page: github.com/skorokithaki... Feel free to use it for your results!

The reason we're releasing this in theatres is because it's incredibly fun to watch in a room full of people. People lose their minds at it. We want people to watch it together. So! Do you run or know somebody who runs an indie theatre? If so, contact us, and we'll let you screen it for free.

I don't know why OpenAI's Python library is always so low-level. You could at least provide a decorator to turn a function into a tool call, guys! I'd write (another) library to do this, but they've already changed the API three times since the last one I wrote.

I can't believe nobody I've talked to about this noticed this bit in Mr Brightside: > Now they're going to bed, and my stomach is sick, and it's all in my head, but she's touching his... chest I laughed out loud when I first heard it, is it just me?

Being on Mounjaro feels I've had an itch all my life and it's finally stopped, it's great. There are, however, lots of people without the itch who'll say "you don't need medication for that, just don't scratch, how hard is it".

I don't know how to reconcile the amazing things people are building with LLMs with my experience of Opus 4.1's "I know you told me this function needs to have four steps but I only implemented the first one because I didn't think the other three were necessary".

I also just discovered Release Please, and it's a fantastic idea: github.com/googleapis/r... It makes a PR with all the changes and change log since the last release, so you can see what you haven't released yet. It also autoincrements the version, so you can release by just merging the PR.

I also really like the board game Just One, so my friend vmasto and I made a simple online version so we can play with friends anywhere we are: justone.stavros.io It's still meant to be played in person, just saves you from carrying cards, markers, etc. The multiplayer bit also reduces dead time.

I also wrote something simple to help me reply to support emails with an LLM: github.com/skorokithaki... The LLM is much more patient than me, so it's a win win. It works very well, if you give it good docs/customer support scripts.

I wanted an easy way to run a sandboxed Claude Code in yolo mode, so I wrote a quick thing: github.com/skorokithaki... It lets you easily define commands to run in Docker containers, passing the env, mounting directories, etc. Pretty useful little command, I think.

I yearn for the years when the users were the customer, not the product. Back then, when we bought something, the company wanted to ensure we got the best for our money. These days we have to be careful not to buy a walled garden where the company will data-mine us for advertiser money. What now?

I'm enjoying Jujutsu SO much. I don't want to go back to git, the only thing I'm missing is a good diff viewer with code highlighting, a la lazygit, and I'm set.

I wrote a simple SaaS with Claude Code over the weekend, and I learnt a few things. Claude Code is great if you keep it on a tight leash, give it specific detail, review what it's doing, and tell it what to change. If you don't do that, it quickly breaks down.

I just tried 4 LLMs to make a simple HTML page that shows a textbox and then shows what you type in full screen. The viewport jumps around and none of them can figure out why. Everyone keeps saying what great results they get vibe-coding, but I can't even get them to fix a bug in 370 lines of HTML.

I had this old rotary phone, so I turned it into a meeting handset. I connect it to my computer via USB, I can hear and speak through the handset, dial numbers with the dialpad, and if I hang up, it exits the meeting. I like it.

I made a new case for my custom TRMNL display, and powered it with a 5 Ah battery, for months of life! I will update my post soon: www.stavros.io/posts/making...

I wanted a TRMNL and I couldn't wait for my pre-order, so I built my own. I also made it so you can easily build one too!: www.stavros.io/posts/making...

I really hate the modern trend of short sessions. With MFA it always takes me at least one minute to log in to each site, and I have to do that multiple times a day, as each session only lasts one day. Just keep me logged in!

Every time I send a support email to @fastmail.com I get a response from a knowledgeable, technical person who is intimately familiar with the product and solves my problem quickly. It's by far the best support I have received, makes me happy to be paying them.

This looks great: www.ftm.eu

AI is great at quickly getting you to "mediocre". For half the people, that's an improvement.

I hate chargebacks, I get zero fraud but have to pay $25 out of pocket every time someone changes their mind about wanting my service and doesn't even bother emailing the (very prominent) address I have on the site to ask for a refund.

The #1 feature of web forums is that I can Google for a solution to my problem and land on the forum. This is why I absolutely detest Discord forums, they are the worst of both worlds: Not real-time (which is why I'm on Discord), not discoverable, and nobody on them bothers to engage.

The more I use iOS, the more I become convinced that Apple has lost its UX-first ethos. I just accidentally swiped up on the middle of my lock screen, and a bunch of old notifications popped up. How was I supposed to have discovered that? iOS is full of "mystery meat" navigation, and it's a smell.

Programming with an LLM is like pulling teeth, it kind of sort of does things, but it does them in a weird way and there are odd bugs. I'd rather use the LLM as a teacher, to learn how to do things, and then program myself, than have to goad it to understand what it did wrong so it can fix it.