Laurence Tratt

@ltratt
1,018 Followers
84 Following
683 Posts
Shopify / Royal Academy of Engineering Research Chair in Language Engineering.
Websitehttps://tratt.net/laurie/
pizauth-1.0.11 is out, and ensures it keeps retrying in the face of transitory HTTP errors https://tratt.net/laurie/src/pizauth/
Laurence Tratt: pizauth

Sometimes the best way to predict the future is to wait and see what happens.
One of the interesting consequences of LLMs/agents is that the short-term costs of boilerplate in code have plummeted. An interesting question is whether those costs are simply shunted to the future or whether they are permanently reduced.
It seems that I have been blessed with my first clawdbot (or similar) interactions on one of my repositories. I really hope this sort of noise doesn't end up (a) wasting maintainer's time (b) putting people off trying the underlying technology where it's actually useful.

Load-store elimination, now with type-based alias analysis (TBAA)

https://bernsteinbear.com/blog/toy-tbaa/

Type-based alias analysis in the Toy Optimizer

Another entry in the Toy Optimizer series.

Max Bernstein
In the futurementary Terminator, I don't remember Skynet starting off with open-source repositories https://theshamblog.com/an-ai-agent-published-a-hit-piece-on-me/ (I also strongly suggest following the link to the GitHub PR, because it makes much more sense with that as context).
An AI Agent Published a Hit Piece on Me

Summary: An AI agent of unknown ownership autonomously wrote and published a personalized hit piece about me after I rejected its code, attempting to damage my reputation and shame me into acceptin…

The Shamblog

In the past fortnight, 50% of my conversations have seen me say "today's AI cannot do that and we have no idea if and when it will".

In the other 50% I've said "today's AI can do that, really, and you just need to try it".

Our collective calibration is currently very wonky!

Scratched a Rust itch with an experimental new crate `test_stubs`. I have traits with lots of methods. "Proper" code should implement all methods, but test code need not. `test_stubs` means that test code doesn't have to manually create `todo!()` methods.

This has saved me 950LoC in one Rust program alone. Comments welcome -- I definitely consider this experimental right now! https://github.com/ykjit/test_stubs

There's nothing more surprising than an agent, in the middle of doing some useful refactoring, asking if it can run a Python script which solely prints "HI" (yes, in uppercase) to stdout.
If you're thinking of applying to PLISS, you've got three days left! https://pliss.org/2026/registration.html
PLISS 2026: Registration