Nate Meyvis
-
Decision model application of the day
I've been thinking about API design in the AI era and about dividing responsibilities between a service and an AI agent calling into that service. I have no firm conclusions here yet, but am still here to report that LLM-friendly APIs enable all sorts of useful things that would be impractical to build into the services themselves. So, I am a trivia fan who: Runs Zippyflash, a spaced-repetition site, which exposes an API; Is a bit behind on his flashcards these days; Wanted to spice up a...
-
A surprisingly great feature
Maintaining bug vigilance and keeping a software-quality mindset more generally tends to make you look like a grump. There are just so many software flaws to notice. Sometimes, though, this sort of vigilance works the other way, and you're the one noticing something great while everyone else is grumbling. At one of my big-company jobs, we used document software (Google Docs-style) that was widely disliked, and understandably so: it was slow, buggy, confusing, and ugly. But it had a wonderful...
-
My Helix fork tutors me
The other day I wrote about how I'm now writing in a WebAssembly fork of Helix. This is working great, even though (i) my vim / Helix knowledge is uneven and (ii) the WebAssembly fork is still buggy. Some extra infrastructure is addressing both of these problems: The editor now saves a rolling history of my N most recent keypresses and recent states of the buffer. Every so often, this information gets sent to a decision model, which determines whether it describes some sort of Helix user...
-
Reading notes: 'My Brilliant Friend'
By Elena Ferrante. People say it is excellent, and it is indeed excellent. Notes (brief, because I don't think my comparative advantage is in analyzing Ferrante): It's nice to read something that makes me feel that a certain kind of literary culture is alive. The mystery of Ferrante's identity, the way this book has similarities to some other big notable recent books (but is also so different from them)... it's a thrill in a way that goes beyond the book itself. I am on Team It Is Very Hard...
-
Programming and fatigue
I find programming much more tiring than it used to be. To judge from what I read, so do many other programmers. But I'm struck by how many of them think the fatigue is a bad thing and a sign they should change their workflows. I've been perceiving it as good, and even comforting. My reasons are simple: if I'm mentally fatigued, it suggests that I've been (i) diligently at work (ii) at things that require human brainpower and judgment. This is comforting because it is compatible with an...
-
What I'm doing with AI these days
A while ago, Trevor McKendrick asked what people are using AI for these days and I answered him. Here's a follow-up. I've forked a WebAssembly fork of Helix, my favorite text editor, and I'm hosting it for my own use on the Web. This is going way better than I expected it would: I can port only the features I need, AI is great at fixing bugs, and latency is low. There's much more to say about this project, but for now: I'm actually using it, it actually works as intended, and it's magical....
-
New problems of deduplication
Programmers spend a lot of time fighting against duplication. Duplicated code is bad, other things equal, because it is inefficient, harder to maintain, harder to improve, much buggier, and much more confusing. Duplicated data is bad, other things equal, for many of the same reasons, but it's arguably even more confusing, because in practice it often takes the form of, e.g., a "master database table" with three different SKU columns that encode subtly different information. Not all...
-
Reading notes: 'Live from New York'
This is an oral history of Saturday Night Live by Tom Shales and James Andrew Miller. I read the original 2002 edition, not the updated 2014 edition. It was a cheap used-book find, and SNL hasn't been much a part of my life since 2002 anyhow. There's a memorable Lorne Michaels quote near the beginning where he says, in effect, that it's hard to make a performance good and that many things end up being not good. Martin Short makes a similar point in Marty, Life is Short (roughly: "that's show...
-
My PRs are getting larger
In the pre-AI world, I was a small-PR1 person, sometimes notoriously so.2 These days I prefer much larger units of code, for many reasons: AI, when used well, has a good chance at making large changes properly. So, to the extent that code review is about ensuring that the code is proper there's less reason to do it at such a very granular level.3 AI allows me to do more, and more thorough, code review before I send it to GitHub or to any other forum for formal review. (Most of this code...
-
Lambda
Here is Patrick McKenzie: I have decided since no one popularized word yet: lambda is the general factor of performance above SOTA LLM baseline. Notes: I am primarily posting this because a term for this is so useful inter- and interpersonally. Wins over replacement is a pretty good analogy for this, but the replacement-level players are becoming arbitrarily numerous and inexpensive. None of this is to say that lambda is the only way to evaluate one's skills. Organizations will use...
-
AI and leisure time
Yesterday I offered, as one of nine reasons why AI wouldn't be leading to noticably more and better software for everyone, that efficiency gains might instead be captured by people making the software: Software teams are capturing a lot of the speedup in the form of more leisure or quasi-leisure time. (To greatly simplify: software teams are offering 2x speedups to their companies. They're then using tools that allow them to work 10x as efficiently, and taking nice long coffee breaks.) It...
- Where is all the new, good software?
- On writing a PDF reader
- Summer reading 2026
- New theories and new software
- Reading notes: 'Will in the World'
- Software wish lists are useful
- Notes on long-running LLM tasks
- Unreasonably effective LLM pattern of the day
- Jevmaxxing