iiinova — where your digital self keeps evolving

Simon @skagedal.tech · 28/09/2026

Fun and engaging review of 2026 in LLMs so far: simonwillison.net/2026/Sep/27/... Although I find it odd that he shrugs of warnings of existential risk as mere marketing ploys. I think it’s pretty clear that they mean what they say, which of course also is a very weird state of the world.

simonwillison.net

2026 in LLMs (so far)

On Friday I gave the closing keynote at the WeAreDevelopers World Congress North America in San Jose. I tied together the key trends from the past year into a chronological …

031

Reposted by Simon

Tyron @tyron.bsky.social · 27/09/2026

Just seen it's Wikipedia's 25th birthday. The greatest website online, and possibly the internet's greatest fruit

19765123

Simon @skagedal.tech · 26/09/2026

Spending most of today trying to track down transactions from my private account for the last year in some kind of attempt to categorize all spending. It's kinda fun to be honest.

010

Simon @skagedal.tech · 26/09/2026

TIL that macOS apps only support ⌘C / ⌘X / ⌘V if they have a menu.

000

Simon @skagedal.tech · 05/05/2026

Assign it to yourself? Although I guess that might more signal “I will eventually review it”, or it’s already automatically assigned to you because codeowner… so yeah, agree 😊

000

Simon @skagedal.tech · 04/05/2026

If so instructed, Claude Code will reach for a proper tool to generate the random id:s.

030

Simon @skagedal.tech · 02/05/2026

No YOU are setting up Prometheus and Grafana to monitor the Kubernetes node that serves your blog that no one reads

010

Simon @skagedal.tech · 01/05/2026

There will be more libraries.

000

Simon @skagedal.tech · 05/01/2026

Very interesting, thanks! Although I'm not on Windows myself :)

000

Simon @skagedal.tech · 05/01/2026

(Which is indeed a relevant amount of faster compared to the shell version, contrary to what I claimed above! I think I might have measured incorrectly. But this all seems to vary quite a bit. But eithe way, the config didn't seem to help.)

000

Simon @skagedal.tech · 05/01/2026

Let's try this one. git config --global core.untrackedCache true With this setting, I've now been running the rust version of my git-dirty-checker under hyperfine to measure run time. It doesn't seem to make much of a difference. Around 210 ms for each run, with or without the option.

100

Simon @skagedal.tech · 05/01/2026

There is also the "untracked cache": git-scm.com/docs/git-upd... I'm not exactly sure how these options relate to, or play along with, each other...

git-scm.com

Git - git-update-index Documentation

100

Simon @skagedal.tech · 05/01/2026

Seems like it'd be nice to have some kind of cache that gets invalidated on file system activity. Git actually has something exactly like this built-in, the fs-monitor: git-scm.com/docs/git-fsm... Is that of any help to our use case?

git-scm.com

Git - git-fsmonitor--daemon Documentation

100

Simon @skagedal.tech · 05/01/2026

it('sums correctly', () => { expect(sum(3, 0)).toEqual(3); }); Seems to work fine!

000

Simon @skagedal.tech · 05/01/2026

And it's not about shelling out to git. It simply takes some time to do a "git status" – it needs to traverse through the directory tree to find any files not in the git index.

200

Simon @skagedal.tech · 05/01/2026

So this approach is still way too slow for the kind of use case I discussed above. And it's not a matter of programming language. For example, I have a Rust version under here: github.com/skagedal/ska... It does the same job in about 0.45 seconds.

github.com

skagedal-tools/git-dirty-checker at main · skagedal/skagedal-tools

A place to put small tools. Contribute to skagedal/skagedal-tools development by creating an account on GitHub.

100

Simon @skagedal.tech · 05/01/2026

Adding parallelism (here using GNU parallel) helps a lot: printf '%s\n' ~/code/*/ | parallel 'git -C {} status --porcelain 2>/dev/null | grep -q . && realpath {}' On the same set of repos, this takes 0.5 seconds.

100

Simon @skagedal.tech · 05/01/2026

To take one step back, here's a brute force solution as a bash one-liner: for dir in ~/code/*/; do git -C "$dir" status --porcelain 2>/dev/null | grep -q . && realpath "$dir"; done This takes about 10 seconds on my example set.

110

Simon @skagedal.tech · 05/01/2026

Imagine the end goal use case something like this: my shell prompt should include the number of dirty repositories under my ~/code directory. Clearly, assuming that I have let's say hundreds of repos, this means caching. But how?

100

Simon @skagedal.tech · 05/01/2026

Here's your programming task. Given a list of git repositories (or worktrees) on your disk, find all that are _dirty_ – that have uncommitted changes, staged or not, or untracked files. The directories where "git status --porcelain" prints nothing. Now do it quickly. No, quicker than that.

100

Simon @skagedal.tech · 07/12/2025

Just made myself a nice little footbath while I start writing another blog post about Node.js. I enjoy the simple pleasures i life.

000

Simon @skagedal.tech · 06/12/2025

I wrote a long post about @pnpm.io's "ignored build script" warnings, why they are there, and what to do about them. skagedal.tech/posts/2025-1...

skagedal.tech

In this post I investigate "ignored build scripts" warnings in pnpm, explain what this refers to, and suggest strategies for dealing with them.

000

Simon @skagedal.tech · 23/11/2025

Wrote a little commenting system. Feel free to kick tires. skagedal.tech/posts/2025-1...

skagedal.tech

Like it was 2003, or something.

000

Simon @skagedal.tech · 22/10/2025

I use git worktrees, or sometimes in other ways have the same code base checked out in two different places. Useful but confusing; sometimes I make some edits in /foo when I meant to make them in /bar. Here's a cool one-liner to move changes over: git -C /foo diff | git -C /bar apply

000

Simon @skagedal.tech · 25/09/2025

The urge to build a tmux/screen type program using this mitchellh.com/writing/libg...

mitchellh.com

Libghostty Is Coming

000

Simon @skagedal.tech · 24/09/2025

Good article by Clearer Thinking on binary thinking. I would like scientific articles to summarize their findings like: “this study aims to study the proposition X. Before conducting the study, we believed that p(X) = … After, we believe that p(X) is ….” www.clearerthinking.org/post/when-is...

clearerthinking.org

When is evidence “sufficient”?

Key Takeaways • The claim that "You should only believe that for which you have sufficient evidence” is popular but it contains an extremely important error. It treats belief as binary: either you bel...

010

Reposted by Simon

Maxim Leyzerovich @round.is · 09/06/2025

sisyphus at daily standup: i had to roll back some changes

435378

Simon @skagedal.tech · 11/06/2025

I also feel like p-zombies totally would report qualia. www.astralcodexten.com/p/p-zombies-...

astralcodexten.com

P-Zombies Would Report Qualia

...

010

Reposted by Simon

Dustin Moskovitz @moskov.goodventures.org · 05/06/2025

Dario Amodei has an important op-ed in the NYT today about the proposed 10-year moratorium on states regulating AI. The moratorium sounds reasonable at first - avoiding a patchwork of state laws - but it's actually quite dangerous given how fast AI is moving. www.nytimes.com/2025/06/05/o... 1/6

nytimes.com

Opinion | Anthropic C.E.O.: Don’t Let A.I. Companies off the Hook

34720

Simon @skagedal.tech · 28/05/2025

I always heard David Bowie sing “we’ve got five years, I pray that’s a lot” instead of “my brain hurts a lot”. Both are fitting.

010

Simon @skagedal.tech · 25/05/2025

Another tech change: I was going to use oauth2-proxy to handle the Google authentication, but I’m realizing it’s probably much easier to just use Auth.js/next-auth…

000

Simon @skagedal.tech · 25/05/2025

In other “welcome to 2025” news, I am now deploying with a Docker container. Still running as a systemd service, ain’t got time to set up a Kubernetes cluster.

100

Simon @skagedal.tech · 25/05/2025

3. And… I am starting to think that React Server Components are really cool. 😎

100

Simon @skagedal.tech · 25/05/2025

2. JSX (TSX). Writing web UI this way is just way better than anything you have in any other programming language (that I know of). All html-as-DSL (like kotlinx.html etc) approaches just feel wrong.

100

Simon @skagedal.tech · 25/05/2025

The things that switched me over sound obvious, but basically: 1. The larger ecosystem. For example, I couldn’t find a decent syntax highlighting system for the JVM. JS ecosystem has such a rich set of tools for that kind of stuff.

100

Simon @skagedal.tech · 25/05/2025

That’s now what’s live on skagedal.tech. A bit too much fighting with Tailwind, and far too much general complexity for a simple blog, but I think it’ll be worthwhile. And I also want to get more deep into that tech stack (also. I want a lot of stuff).

skagedal.tech

Thoughts on software development, technology and the future.

010

Simon @skagedal.tech · 25/05/2025

Well, its so happened that I scrapped this whole approach and instead rewrote things again in Next.js 😄

000

Simon @skagedal.tech · 24/05/2025

Is something missing on the code line under “// Interact with UI”? Shouldn’t there be a `user.click(button)` or such?

000

Simon @skagedal.tech · 20/05/2025

I do sometimes like to drop the .sh extension on shell scripts because “it’s cleaner”: tools should be able to identify it using shebang for syntax highlighting etc, I don’t need to expose the implementation detail just like with any executable, etc. But sometimes I wonder what I do with my life.

000

Simon @skagedal.tech · 18/05/2025

It really is a very different claim from those others. The claim isn’t that “non-programmers will be able to make software”, it is that there will be a new kind of programmer, whose brain is silicon-based rather than meat-based.

000

Simon @skagedal.tech · 14/05/2025

What if you say something that gets interpreted as ”> If that also isn’t properly escaped, you could affect the xhtml payload in unexpected ways?

000

Simon @skagedal.tech · 14/05/2025

Very interesting! Although I don’t quite understand your argument that this couldn’t be a security vulnerability. Of course, the invalid XHTML is refused. But the issue is that it’s there in the first place, rather than the ampersand having been properly escaped. >>

100

Simon @skagedal.tech · 05/05/2025

They’re most likely potheads.

010

Simon @skagedal.tech · 27/04/2025

Alternative answer: do just as described in your scenario, but pay close attention to, and find deep meaning in, every moment. No time wasted, perfectly optimized.

000