Location: Switzerland
  Remote: Yes (hybrid or on-site for a good fit)
  Willing to relocate: If it's a a good fit
  Technologies: LLMs, TypeScript, Bun, React, Remix, Next, Node, Django, Python, Tailwind, SQLite, OCaml, Nix
  Résumé/CV: https://www.linkedin.com/in/josef-erben-90a002b1/
  Email: josef@erben.systems
  Blog: https://www.joseferben.com/
  Github: https://github.com/joseferben

I am currently bootstrapping a product but I am open for employment if it's a great fit. I have over 10 years of experience as engineer and founder. As an engineer I've worked in finance, research (economics & health care), education, real estate, marketing and dev infra.

Recently, I’ve been using Typescript, React, Bun/Node, Django and Python to build web apps and agentic workflows on top of foundation models. I have some experience with fine-tuning and with reinforcement learning, so I’m also interested in work that is closer to research.

I have my own LLC and I am also open to contract work.

If you want to drive Claude Code or OpenAI Codex with code comments this is for you. Make sure to set up a good check command that takes files as arguments, such as eslint. The agent fixes issues in the changed files first, making the feedback loop tight.

congrats on the launch! better-auth is a joy to use. great dx with the type-safe plugin system.

the quality of the lib and the docs is really high, kinda crazy you just launched!

for me i figured out it’s about the body. it’s ok to be lifted up from the body into the thinking mind but i “owe” my body to spend some time there as well.

sometimes all it takes is sitting 20min in the morning just observing sensations in my body, and saying good morning to various organs haha. sounds silly but creates a solid foundation for my day.

> layout, "server-only", accidentally bundling server code

the main issue i wanted to highlight is that things are not caught at type-check time. maybe vercel could provide eslint plugins? this could codify a lot of the caveats

> There's no way to get around thinking about AuthN/Z with any server APIs

very true! maybe there is a way to make the default secure somehow by making it awkward to publicly expose a POST route by accident. awesome that next.js 15 is doing things in that regard.

thanks for addressing the points leerob, shows the great governance mentioned in the end of the post :)

for example:

imo it's easier to spot the authn/authz issue here

export default function handler( req: NextApiRequest, res: NextApiResponse<ResponseData> ) { // updating user details without authn/authz }

than here

"use server"

export async function updateUser(email: string, age: number) { // updating user details without authn/authz }

without having read any next docs. engineers got trained to consider authn/authz the moment they see "http route/endpoint", maybe being more explicit about what happens in the api would help?

or next could provide an authn api for server actions that authn providers would implement?

after building 5 apps with next.js 14 this is our review.

tl;dr

- Next.js 14 is fast, but App Router caveats increase cognitive load

- Server Actions are awesome, once secured

- Everyone is using Next.js, this is its greatest strength

- Vercel as hosting platform is a joy to use and they save us tons of time

while working on a few slack apps and llm agents i couldn't help but notice how chat platforms like slack could be to agents what message brokers (kafka, rabbitmq, sqs) are to services.

what do you think?

excellent points

@simonw what is the token/s like on your 64gb m2 mbp?

With MLX:

    Prompt: 49 tokens, 95.691 tokens-per-sec
    Generation: 723 tokens, 10.016 tokens-per-sec
    Peak memory: 32.685 GB

so quite usable, thanks!

looks good! how does it compare to the core of vercel ai sdk?