Ontario, Canada
- Tweets 12,315
- Following 1,061
- Followers 8,479
- Likes 53,848
> See Rust benchmarks on X
> Rewrite your SaaS in a language you don't know
> Benchmark it
> 30x more requests per second, nice
> Still 6 users
> Ask AI to add a new feature
> 47 files changed
> You can no longer review the code, no biggie
> Your production environment now depends on your ability to explain yourself in perfect precision
> Realize your actual programming language is now English
> Every sentence has undefined behaviour
> The compiler responds with "Absolutely"
> Push right to master
> Discover what you meant in production
22
14
190
6,953
Steve Bauman retweeted
okay, the time has finally come 👀
One of the most requested features for Laravel AI SDK is HERE. 🔥
Your agents can now have Skills. 🧠
Agent skills support dropped in AI SDK v1.1.0
github.com/laravel/ai/releas…
14
17
206
24,558
Introducing 'Checkpoints' for Operations v1.2.0 🎉
You can now save progress in your operations to resume from it after a failure
Add the HasCheckpoints trait to your Operation and persist information you need to skip already processed rows:
3
3
43
1,780
We've all encountered this issue where a backfill query doesn't run as quickly as expected, isn't using the proper database indexes as you expect it to in production, or fails half-way on an unforeseen edge case
This allows you (or an exception) to bail hard on an operation and then start-up where it left off vs from scratch each time
216
I've been seeing an immense number of posts about using AI to rewrite codebases in Rust for better performance
What worries me is how easily "the LLM can understand it" becomes a justification for code that the people maintaining it can barely follow
At that point, we're accepting complexity because we assume AI will always be there to explain it, change it, and fix whatever breaks. The code gets faster, but our understanding of our own systems gets worse
Aren't we optimizing for the wrong thing?
With tools this capable, shouldn't we be spending the tokens to make the code expressive? Choosing better names? Simplifying the structure/architecture? Making the intent obvious to the next person who opens the file?
Being able to ask an LLM what something does shouldn't become an excuse to stop making that clear in the code itself. We still need to review changes, question assumptions, and understand why something went wrong
It reminds me of the Tower of Babel - we keep building higher, but the people doing the building lose the shared language needed to understand what they're working on. The LLM becomes the translator we depend on to keep going
Some performance gains are worth the complexity. But human understanding should be part of that calculation
If AI gives us the ability to make our codebases easier for humans to understand, why are we so willing to give that understanding away?
16
5
63
3,242
Just released Operations v1.1.0
You can now run individual operations by filename, making it easy to test your one-time deployment tasks with Pest
Create your data, run the operation, and assert the result:
3
2
34
2,199
I've just released 'Operations' for Laravel 🎉
A small and simple package for one-time deployment tasks, like backfills and seeders
Write your handle() method and run it during deployment. Completed operations are recorded and skipped on future runs:
github.com/DirectoryTree/Ope…
12
10
151
7,603
Need to deliberately repeat completed work? `operations:forget` removes its completion record after confirmation, so you can run it again if necessary:
1
2
363
I want a vibe coded operating system as much as I want any other vibe coded product
Which is not at all
2
29
1,691
After 1000s of messages, 10s of chats, millions of tokens, hundreds of dollars in usage, ImapEngine v2.0 is nearly complete
You deserve a well crafted and beautiful library, and I will deliver
2
48
1,900
I'm struggling to see this on a daily basis. I'm constantly adjusting/correcting/tweaking/clarifying architecture on frontier models on max thinking
Maybe I'm doing it wrong? 🤷♂️
I understand the appeal in trying to find the first plausible fortress in our retreat from writing code, but if you think it's "architecture", I have bad news for you. The models are also very good at that.
7
38
6,084
I've just released Runnable: for Laravel 🎉
A small package for running your action and query classes through Laravel's container, with a simple API for faking their results in your tests
Add the trait, write your `handle()` method, and call `::run()`
github.com/directorytree/run…
5
8
99
5,659
There's three ways to run:
- The `Runnable` trait
- The `Run` facade, or;
- The `run()` helper
Each calls `handle()` through the same runner and returns its result
The facade and helper also work with classes that don't use the trait:
1
3
520
I built Runnable because I kept defining the same conventions for running and testing focused classes in each project
I wanted a small bit of structure that felt right at home to Laravel developers, with easy faking, while leaving how you organize your application up to you
I hope you find it as useful in your projects as I have!
6
370