Brian Slesinsky
skybrian.bsky.social
Brian Slesinsky
@skybrian.bsky.social
Retired software engineer, amateur accordionist. Other accounts:

https://mastodon.social/@skybrian

https://tildes.net/user/skybrian
Reposted by Brian Slesinsky
Genuinely very impressed by the SVG of a pelican riding a bicycle I just got out of Google's new Gemini 3 Deep Think model simonwillison.net/2026/Feb/12/...
Gemini 3 Deep Think
New from Google. They say it's "built to push the frontier of intelligence and solve modern challenges across science, research, and engineering". It drew me a really good SVG of …
simonwillison.net
February 12, 2026 at 6:16 PM
Reposted by Brian Slesinsky
I don’t really know what I expect in posting this, but maybe some people see it and can learn from it.

I’ll call it “lessons from an unwilling immigration attorney.”

1/
February 11, 2026 at 9:56 PM
Reposted by Brian Slesinsky
exe.dev is still behind a credit card for now while we develop some more techniques for dealing with people who abuse systems (even a credit card doesn't entirely work). But if you are a human who wants an invite code to skip the CC for a month, happy to DM you one.
February 11, 2026 at 12:43 AM
Reposted by Brian Slesinsky
LLMs are kind of like sails in that left free flowing they're completely useless but tightly bound and directed they can dramatically accelerate your progress
February 5, 2026 at 8:53 PM
Reposted by Brian Slesinsky
Wrote blog.exe.dev/expensively-... to dig into how cache reads costs dominate LLM agent conversations. Several visualizations and one terrible pun included!
February 3, 2026 at 5:28 PM
Reposted by Brian Slesinsky
In Medieval Europe ships called cogs revolutionised shipping simply by their size. A cog named Svælget 2 was recently found off the coast of Copenhagen: it’s 28 metres from bow to stern and, preserved under sand, its rigging is still intact. buff.ly/tqiJIJQ
#ShareGoodNewsToo
Archaeologists find a supersized medieval shipwreck in Denmark
The sunken ship reveals that the medieval European economy was growing fast.
buff.ly
February 2, 2026 at 1:50 PM
Reposted by Brian Slesinsky
February 1, 2026 at 11:48 PM
If you could give everyone posting on Bluesky a prompt, what would it be?
February 1, 2026 at 4:19 AM
Reposted by Brian Slesinsky
> tailscale is the secure way to connect all your computers!

> i've used tailscale to connect my molty to all my computers and internal services and gave it all my keys

oh no, not like that
January 30, 2026 at 10:11 PM
I see *curation* as a major counterforce to slop. Yes, AI spreads slop like kudzu, but not when there's a committed gardener. AI is also an under-appreciated power tool for getting rid of slop in whatever garden you choose to take care of.

If your garden is overrun, maybe use better tools?
Your codebase is the prompt - exe.dev blog
When Claude writes code you don't like, ask what it saw, not what it did.
blog.exe.dev
January 28, 2026 at 3:53 PM
Reposted by Brian Slesinsky
How hard can it be to build a browser from scratch for three platforms anyways?

Apparently 20K lines of code and ~70 hours from first commit to last.

emsh.cat/one-human-on...

#llm #llms #ai #codex #openai
January 27, 2026 at 1:26 PM
1. Create a new conversation
2. Paste in "Could you review design doc 019. Is it ready to implement? Any suggestions?"
3. It makes several good suggestions, so I ask it to fix them.
4. Goto 1.

I guess that's a Ralph Wiggum loop, done by hand.
January 27, 2026 at 6:25 AM
Reposted by Brian Slesinsky
Was editing and noticed that if you rotate-left "those" you get "ethos". Was curious how many of these are in English. Big dicts have a lot of junk, and most endings in "s" are for plurals. Here's a good list of rotwords.
January 25, 2026 at 4:17 PM
@spacecowboy17.bsky.social The "For You" feed was pretty good for me earlier today (lots of developer posts) and now it's all ordinary political stuff. I'm wondering if there's an effect where if you check it too often, it runs out of good posts and you just get random stuff? Or is it time of day?
January 23, 2026 at 3:43 AM
Reposted by Brian Slesinsky
Whenever I read discourse on AI energy/water use that focuses on the "median query," I can't help but feel misled. Coding agents like Claude Code send hundreds of longer-than-median queries every session, and I run dozens of sessions a day.

On my blog: www.simonpcouch.com/blog/2026-01...
January 20, 2026 at 2:38 PM
My hobby programming project is turning into a bureaucracy. I asked the coding agent to create a design doc directory and a template for writing new design docs.
January 18, 2026 at 9:07 PM
Reposted by Brian Slesinsky
Imagine writing a pure C, dependency free, fast enough Flux 2 Klein 4B inference library. In a weekend. Thanks to AI writing the code, and the human steering it towards the right direction.

github.com/antirez/flux...
github.com
January 18, 2026 at 5:58 PM
Did a minor release of #repeatTest to add a few things now that I'm actually using in a couple of projects. Or really, my coding agent is. So I wrote better docs for coding agents, among other things.

#testing #deno #typescript #javascript

jsr.io/@skybrian/re...
@skybrian/repeat-test - JSR
@skybrian/repeat-test on JSR: new, experimental library for writing property tests
jsr.io
January 18, 2026 at 1:02 AM
Current status: two hungry ghosts in jars. #multiclauding exe.dev
*Memes are showing up. #multiclauding
January 17, 2026 at 12:20 AM
I rarely like political posts, but this one goes beyond stating an opinion and makes an actual argument.
Let’s talk about abolishing ICE.

Very few people who say “abolish ICE” mean “abandon all immigration enforcement.” Rather, they mean “the structure of immigration enforcement is irretrievably broken and needs to be reorganized, just like we did 23 years ago when we created ICE.”
/1
January 14, 2026 at 9:26 PM
Reposted by Brian Slesinsky
Until C# all code was blurry and really hard to read, that's why all programmers needed glasses
Until Objective C, all code was pointless, random. The big innovation was having a goal with it
January 14, 2026 at 1:29 AM
Here's a cool idea: ask AI to implement a spec multiple ways, then run them in parallel to see what the diffs are. Either the diffs are allowable, an AI screwed up, or it's a spec bug.

I think they wrote multiple implementations for the space shuttle, but now we have coding agents so it's easier.
Differential spec analysis
Differential techniques are chronically underappreciated in software. And they pair incredibly well with coding agents. Agents are now good enough that differential analysis can even be used to refine...
commaok.xyz
January 14, 2026 at 9:06 PM
Reposted by Brian Slesinsky
If you can substitute "hungry ghost trapped in a jar" for "AI" in a sentence it's probably a valid use case for LLMs. Take "I have a bunch of hungry ghosts in jars, they mainly write SQL queries for me". Sure. Reasonable use case.

"My girlfriend is a hungry ghost I trapped in a jar"? No. Deranged.
August 13, 2025 at 12:56 AM
A side effect of using a good coding agent is that now I want to fix *everything* because I imagine it will be easy. I just need to get a spare moment.
January 11, 2026 at 4:56 AM
It's quite nice when a coding assistant can write a bug report for you, suggest a workaround, and then fix the bad record in its own sqlite database :-)
Empty assistant message saved to conversation causes unrecoverable 400 error · Issue #3 · boldsoftware/shelley
I asked Shelley to write a bug report - skybrian Shelley sometimes saves an assistant message with empty content ("Content":[]) to the conversation database. When this happens, the conversation bec...
github.com
January 10, 2026 at 11:53 PM