I think languages like Go might thrive:
- Type safety
- Little sophistication in terms of language features
- Fast compile times
I have been having good success with Rust. I wish it compiled faster, but I understand why it doesn't.
My TypeScript experience has improved using bun instead of node.
I think so. On the other hand, I think there might motivation for languages and tooling that are optimised to work better with agents: built-in safety, stronger correctness, tight feedback loops, compact outputs to avoid wasting tokens, etc.
Hope you don't mind a bit of a shameless plug, but just sharing in case it is of any interest at all: we have validation+review of public skills and it flagged a couple of small suggestions
tessl.io/registry/ski...
A @flipper83.bsky.social le van a pitar los oídos
We have been focusing on other ecosystems first, but DM me if you are interested in trying it out for key Ruby gems 😄
At tessl we are generating docs for libraries to help agents perform better. We found that our generated docs provided a similar boost in performance to giving the agent the source code, but it took less time and turns to achieve the results.
tessl.io/blog/propose...
I've been playing with unsupervised agents too. I had an instance where the agent not only checked the source code for a library, it found a bug, forked the library, fixed the bug, and pointed the deps to the fork 🤯
Some agents will do this automatically for certain ecosystems. They dive into the node_modules or the cargo crates cache to get to the source code, or even go to GitHub directly.
Isn't it going to be interesting once LLMs start training on all this data? 😂
A GitHub workflow triggered on push that adds tangled as a remote and pushes the changes there?
Full stack se queda corto :)
I’d love to buy a coffee and share some notes if you’re up for it.
We’ve been focusing more on how to speed up execution (increasing throughput and reducing lead time) so far. Your examples seem more related to streamlining work after shipping, which makes perfect sense.
That’s great! I’m curious to know how you’ve adapted your ways of working in light of this.
A few months ago I realized that, given our journey with coding agents, my intuitions about WoW best practices should be reconsidered. It's interesting to see how different people are adapting.
Instead of just doing a gist, I ended up spending some time with CC to set-up a config pack so anybody can install it and get started: lazyjj.dev
Let me know what you think
OMG, I haven't listened to Ska-p in so many years. What a blast from the past 😅
I created a `jj clresolve [change_id]` that opens CC with a prompt to solve the conflict in plan mode.
It has been very helpful.
gist.github.com/ernesto-jime...
Totally!
Another angle:
* My parents learned by memorising because finding information was time consuming
* Mine involved knowing how to effectively find the info, but I still had to consume and distill
* Now it is going to be verifying and editing
And that's assuming we don't solve hallucinations
Es el reajuste más grande que he visto en cómo trabajar en toda mi carrera.
Muchas de las maneras de trabajo que tenemos se basan en el hecho de que producir código es la parte cara, sufrida, y difícil de cambiar.
En muchos ámbitos ese ya no va a ser el caso.
A muchos que se nos había acabado la oportunidad de cacharrear al tener peques, CC nos la ha devuelto.
En el curro he tenido algún día de abrir 15 pull requests en un día. Ahora tenemos un agente haciéndonos las reviews y, si no es un cambio delicado, mergeamos si el agente nos da el 👍
Otro ejemplo, en el cole de Bruno estaban haciendo una subasta online para recoger fondos y yo tenía curiosidad por la actividad.
En una hora o dos CC me había hecho un servidor que descargaba la web, extraía las pujas, y me enseña quién pujaba por qué en tiempo real.
Yo convertí un juego que hicimos @ladybenko.net y yo para una competición hace muchos años de un lenguaje de videojuegos a una web sin picar una línea de código. Y me llevó una tarde.
Yo encantado 😄
Si no lo has probado todavía, te recomiendo que le eches un vistazo. Puedes probarlo "full-vibes":
1. Crea un directorio nuevo
2. Shift+tab,tab para decirle que aceptas el código que quiera escribir sin revisarlo
3. Pídele haga alguna aplicación pequeña que te gustara tener
Cuánto tiempo tienes? 😅
Hay mucha funcionalidad que se solapa, pero yo uso Claude Code de forma muy distinta a como usaba Cursor y Copilot.
La analogía mas sencilla es que con Cursor yo hacía pair programming con el agente. Con Claude Code delego tareas.
The divergent syntax sounds like it might to save me so much pain in the ocasional situations where I get unintentional divergences! ❤️
Congrats on the first contributions!
Same here but with MacOS and Windows, because I got a gaming PC and thought I would want to use WSL more than I actually have.
I do indeed! but only through my `clstart` and `clstop` aliases which are tied to my tmux set-up too.
This one I could gist pretty easily 😊 gist.github.com/ernesto-jime...
As a bonus, it also includes the hooks I use to have my mac talk to me when Claude Code is waiting for me.
Not right now, but I've got a couple of flights next week. I'll see if I can find some time to clean things up and share :)
Te compras el dominio y que redirija al un cal.com para las vistas 😂
Me alegro que lo estés disfrutando tanto! 😄
I got spoiled with the 2x limit on my Max plan over Christmas 😅
You can try doing this for yourself using git-ai too: usegitai.com
They store the prompts in Git notes within your repo.
