@carllerche

Creator of @tokio_rs. Building Toasty, an ORM for Rust (https://nitter.cf/t.co/1Fm2upgNyK). Prev: Cargo, Ruby's Bundler, and others. Ruby on Rails core team alum.

Portland, OR
Joined April 2007
Carl Lerche retweeted
Disagree just as high level languages help us code better - there's less to think about, it will be the same for AI. Why reach for ASM when you can use Rust - there's so much more that can go wrong. By this logic AI will also forego testing because they have so much intelligence to just reason through it.
Rust is a good prompt compilation target for the moment, but so is C++. And soon assembler. Then microcode. Myopic to think we're going to stop the agentic drill bit until it reaches computing bedrock.
57
52
15
886
86,451
I am also acutely aware that frameworks like Topcoat may only be relevant for 5 more years, give and take. I predict there will be far fewer “apps” vs just exposing data and actions to AI. 90% of my app interactions today I wish I could just go via ChatGPT instead.
Topcoat is pushing the boundary of server applications with Rust. Given the discussion falling out from @dhh's keynote, I thought I would write a bit about why I'm working on Topcoat and the cool features that were recently shipped: tokio.rs/blog/2026-09-24-top…
5
2
49
2,887
I also expect that chat is just the medium used today.
400
This is possible, but I am skeptical. Unless LLMs become cheaper and faster than running a compiler, a deterministic translation from higher-level abstraction to microcode will be the way. I could see a more efficient binary "language" for AI, but still with a compilation step.
Rust is a good prompt compilation target for the moment, but so is C++. And soon assembler. Then microcode. Myopic to think we're going to stop the agentic drill bit until it reaches computing bedrock.
10
3
56
5,432
There is also the question of ecosystem. LLMs are cheap, but not free. Any future AI target would greatly benefit from being compatible with an existing, large, ecosystem.
9
496
This is my point! Everyone says AI agents work best with their language. This means that AI agents are equally good at generating code in any language. So, you might as well pick the programming language that gets you the fastest, tightest tail latency, lowest memory end product
36
2
6
116
16,038
Carl Lerche retweeted
Using Turso database via Tokio's Toasty ORM is insanely good
1
1
3
1,965
Topcoat is pushing the boundary of server applications with Rust. Given the discussion falling out from @dhh's keynote, I thought I would write a bit about why I'm working on Topcoat and the cool features that were recently shipped: tokio.rs/blog/2026-09-24-top…
6
22
8
172
23,973
Topcoat v0.9, the batteries-included, full-stack Rust web framework, lets you push live updates from Rust straight to the browser with the live! view macro. Try it out now!
2
5
30
1,235
AI is ushering in a new era of abundance... including shitposting 😂
4
15
1,088
...present party included.
3
378
Carl Lerche retweeted
Critical high throughput services moving to Rust is going to save billions in CPU and memory costs, across the industry.
6
3
1
28
2,513
Come on @dhh. I know Rust is no Ruby, but the ugliest language ever created? There has to be uglier.
60
10
5
368
60,915
That is from Topcoat: github.com/tokio-rs/topcoat. Trying to capture some of the magic that I felt building apps with Rails, but with Rust (and really really fast).
5
1
38
6,208
Wait... did he actually say that in his keynote?
Scoop: @dhh will be using Rust to build web apps in 5 years. You heard it here first! 😅
7
2
56
8,253
This is what I tell Rust peeps too. String strings... str strs... CStr CStrs... OsStr OsStrs... and OsString OsStrings.
Replying to @valigo
It’s actually very easy. std::map maps. std::flat_map flat maps. std::unordered_map maps unorderedly. No need to complicate things.
4
1
1
51
9,971
Other parts of my flow. Fire off large attempts at features, see what the LLM does, then discard it and fire off smaller change prompts to build up the abstractions needed for the full feature. Also, I repeatedly ask the LLM to reduce a patch size until it stops helping.
Replying to @timClicks
Same way I guarded against slop when accepting contributions in OSS projects. Focus on enforcing small, modular components and abstractions, relax on the details within each component. If the "slop" is contained, it can always be fixed later. LLMs are 100% making me go faster.
14
1,665
Carl Lerche retweeted
Replying to @timClicks
Same way I guarded against slop when accepting contributions in OSS projects. Focus on enforcing small, modular components and abstractions, relax on the details within each component. If the "slop" is contained, it can always be fixed later. LLMs are 100% making me go faster.
1
3
1
47
3,044
Carl Lerche retweeted
I merged a change that should reduce the size of your Rust `target` directory by ~20% during development. It turns out the compiler was retaining an extra completed snapshot for every crate that was incrementally recompiled.
71
49
15
1,416
51,671