Screenshot of several images from my new blog post
Blog post: I've been working on a guide to SDF font rendering, and I kept getting stuck. The blog post is about the things that didn't work www.redblobgames.com/blog/2026-02...
@interjectedfuture.com
Tech Zine Issue 1: LLM System Eval https://forestfriends.tech Local-first/Reactive Programming ⁙ LLM system evals ⁙ Startup lessons ⁙ Game design quips. Longform: https://interjectedfuture.com Podcast: https://www.youtube.com/@techniumpod
Screenshot of several images from my new blog post
Blog post: I've been working on a guide to SDF font rendering, and I kept getting stuck. The blog post is about the things that didn't work www.redblobgames.com/blog/2026-02...
File this under more of "The future is weird, for anyone paying attention."
x.com/dioscuri/st...
I wish he called it a Familiar or a Summon.
There's a direct line from Claude Shannon to Lobster Appendages. Hope etymologists are taking note.
x.com/karpathy/st...
By forcing me to log in, Perplexity just...makes me stop using it.
Is writing outsourcing our thinking? Socrates seemed to think so.
Is there any talk about clawing it back with upcoming classes of Congress? Or they've just kinda let it go, given how often we're just in low-key wars?
Honest question: I thought only congress had the power to declare war. I know that’s been weakened in recent decades. Have they given up on it altogether?
Ah. I assume they have a base implementation that adheres to the protocol, but with these forward looking variants on top, which other peers ignore if they don't understand it, but it doesn't break the protocol?
I don't have a lot of context, so bear with my dumb question: are buckets experimental, or are they something that's landed on AtProto?
Anatomy of an LLM connection string llm://app-name:password@api.openai.com/gpt-5.2?temp=0.7&max_tokens=1500 The scheme is llm://. The host is the provider’s API base URL. The path is the model name. And query parameters handle all the runtime options that usually clutter your code.
Being a bit of a URL nerd, I love this idea
danlevy.net/llm-connecti...
Is there an OpenClaw for managing your Kubernetes cluster? Or is the OpenClaw of Kubernetes management just OpenClaw?
The future is so weird. It's hard to express to people not paying attention as to how weird it really is.
Selling to people though, I think will still have humans in the loop for a long while.
A follow on consequence of this is that people will try to figure out how to do sales to agents. And subsequently, how to get agents to sell to agents.
Agent-led sales.
x.com/snowmaker/s...
So if big tech uses industrial processes like agentic engineering to code, and most of agentic engineering is making architectural decisions in specs and Q&A; I would think that interviews would now focus mostly on systems design questions.
One of the big step change difference between Opus 4.5 and 4.6 for me was....that it could one-shot kubernetes configuration of ingress for ActivityPub on my blog.
Sometimes, the scar tissue of doing something sucky (like dev ops) makes me procrastinate on doing it, when in reality, it's much quicker with agents now.
I bet you 150k token context window looks like a million token context window to those that write APL.
With so many more non-traditional coders being turned onto coding via agents, this is one of those rare demographic and platform shifts.
So if any APL/Datalog/Haskell fans want to indoctrinate new programmers, now is the time. Notation as a Tool for Thought as a way to prompt.
If programming is theory building, then lots of people aren't building any theory on the technical shape of the problem, because the tools are inadequate during review. Not only are we bottlenecked there, but we're also becoming detached.
I have a mildly cold take
bsky.app/profile/inte...
Just learned a new term "Info Civics". System evals for certain public services should be public, like we have with the court of law.
github.com/pfrazee/inf...
I've been watching AtProto from afar. So far, I've liked what I see, and when it comes to decentralization, I really like how pragmatism is in the list of considerations.
www.pfrazee.com/blog/practi...
I just wrote about this. Haha.
bsky.app/profile/inte...
It'll provide a long tail of software never seen before, which will spring forth new industries. It'll have long-reaching second-order effects like the aforementioned drive-thrus and highways, now that intelligence is on tap.
For everything else one has midwit tastes and needs for, we'll probably lean on industrially produced software using agentic software processes. It'll democratize access to using computers, not just as a consumer, but as a producer. Or even just to let us focus higher on the meta
There will still be artisanal coders in the future, either for luxury or status signaling products. And they will provide an experience far better beyond what industrial software could provide, *if* you're sensitive to it--an otaku for that sort of thing.
In the long term, new ways of building will be around the process of managing the agent build process and understanding the mathematical nature of composition, in the same way current industrial manufacturing innovation is around process and material science.
So I think coding by hand is important in the near term for finding new forms and ways of building, since LLMs currently have a hard time drawing the right system boundaries and abstracting beyond what's shown in the current training data.