Sharing my experiments with AI. Personal stream of consciousness.
Joined May 2011
- Tweets386
- Following163
- Followers337
- Likes641
Hi guys, please roast my cherry blossom. 🌸 What should I do to improve it?
My past vibe-coded experiments tried to one-shot too many things at the same time. I've learned my lesson and I'm trying out a new idea, one small step at a time.
Made with Fable 5.2:
4
4
149
Claude's "you should know" plugin checks changes for issues you might not know about.
It's caught issues for me many times and helps me think about the PR more clearly.
Enable it by installing "cc-plugin-you-should-know" in /plugins.
32
this is maybe a dumb question, but
why does X only notify for some replies? Just noticed I had replies on older posts that I never saw.
1
2
79
Kevin retweeted
Today marks 5 years of @TerraformIndies. We've come a long way, and I'm grateful to be building in the excellent company of our team and investors.
Some actual content so @lulumeservey doesn't hate me...
Not all hardware start ups face this issue but for us our major risk has always been about controlling capex. There's no market risk for oil and there's no tech risk for synthetic chemistry that has been around for a century, but producing oil for $2000/barrel earns a participation trophy.
Rather than spend a fortune building functional prototypes and then another fortune trying to figure out how to suck out 95% of the cost, I placed firm upper bounds on allowable BoM costs for each system, then set out to figure out how to get it done. Some of the early prototypes didn't work that well but we kept iterating and ultimately every cent and, more importantly, every second was spent figuring out how to do it the right way. The way that accelerates great value into the hands of our customers.
History is unfortunately littered with expensive tech demos that spent millions teaching someone the wrong way to do something. At this point it seems unlikely that Terraform will die but if we do, it won't be because we spent our money on insurmountable tech debt.
The reason I'm sharing this is because it's always tempting to take the shortcut. To outsource some piece of mission critical kit or design, or to drop some gold plated component in that removes more value than it adds. Don't! The whole point is to spend your limited time on this Earth attacking the really hard problems. The things you want are on the other side of discomfort.
Here's to the next five years!
We've all been doing PRs and code reviews wrong.
I read Anthropic's blog on how the Claude team does code review, and a few things became obvious:
* PRs should be smaller and more mechanical, so they can be more easily reviewed and merged.
* Reviews should run in a fresh context. After 3 rounds, reconsider the approach. One PR looped through more than 50 reviews before the author stopped it.
* Pre-existing issues, style issues, code cleanups, nitpicks, and things like that should not block the PR. They can go in their own PRs if needed.
What I'm currently doing:
* I use a boss to intermediate between the implementor and the QA.
* Findings and decisions are logged in a plan file that all agents can read.
* I often add a cross-family check.
Recommended reading: 👇
1
57
"How the Claude Code team does code review": claude.dev/blog/solving-code…
33
Kevin retweeted
New blog, in which I describe which layers of the AI stack are getting substituted the fastest and slowest.
blog.caseyhandmer.com/2026/1…
Enjoy!
Some shots from a side project I've been working on: a view of the Earth on the nighttime side, and the moon. You can faintly see the glow of the city lights as well as that of the stars, and faint airglow around the Earth.
The moon is lit up by the sun and greatly outshines the stars, which is why the sky is black there.
The idea behind this is to be able to walk around this in AR and feel the vast scale between objects (and also to push AIs to the limit!)
I've discovered that the visuals and the physics behind this are a deep, deep rabbit hole and the images below are only an approximation. There is a profound beauty in the universe when you follow the math. Full respect to the people who work in this space every day.
1
50
did you know you can give your mac caffeine to keep it awake?
caffeinate -dims
d: keep the display from sleeping
i: keep the system from sleeping when idle
m: probably not needed with SSDs, but keeps the disk from sleeping when idle
s: prevents the system from sleeping at all, but this only works when plugged in and you don't close the lid.
Both Claude (by default) and Codex (with an experimental flag) can do a light version of this internally. You can also tell them to run caffeinate directly when you:
- want to keep browser tabs animating
- need to run GPU tests
- want to keep the display on
Ask your agent if your project would benefit from caffeinate. They'll see if it makes sense and which options you need, and when to give your mac a break from the caffeine.
4
2
162
Even if you don’t burn the brightest, even if you find yourself in the void, your light still matters. Your light adds to the glow of a vast, starry universe.
1
52
Anyone else wish we could go back to the old split screen multitasking on iPad? The freeform windows drive me up the wall.
1
2
98
We're all sick at home today, so I've been working on this evening lantern garden. In one part of the garden, there's a farm with cows, horses, a pig, cats, chickens, and ducks, and at night, the animals lie down and go to sleep.
Music: "A Night on the Farm", in G major, by Claude (sound on)
"At dusk, the chestnut mare comes to the fence to say goodnight to the cat. One by one, the farm settles: the ducks on the moonlit pond, the hens in the coop, a kitten curled up against her friend. The lamps go out. And just before morning, everyone wakes."
This started out as a one-shot experiment. I have some work to do on the visuals (claude binded too hard to my previous "obby game" experiments), but so far I really like claude's choices for the music and sound.
3
4
315