Cillian Barron

@ScriptedAlchemy For real. Just go all in on remote sessions/VMs and let me manage it all from my phone

The fact that there’s no codex iOS app gives me very little faith in AI labs “getting it” Electron app is cute and all. Can you please just build tools that developers actually need and want. Why the fuck do I have to use a shit web page in safari for a cloud environment or be glued to my computer. You sell us this “amazing productivity” yet fight tooth and nail to avoid giving us anything that would actually boost productivity. It’s so painful to watch OpenAI fumble the ball with codex. The model is unbeatable, but you build the harness like it’s a bunch of 9-5 ticket draggers who have no passion in building software. Hire people who actually use codex 16 hours a day to build it ffs.

@thisismahmoud @RileyRalmuto @Railway Have to say the railway skill + cli combo is absolutely unreal for devex. I’ve got a prototype app with 3 services/db/redis/env vars stood up on railway and I think I’ve looked at the dashboard once. Codex was able to debug and fix a cors error and everything. So so good 👏

@RileyRalmuto Saw @Railway getting mentioned. Happy to answer any questions 🫡

slowly but surely, this is going to be the experience and expressed sentiment of virtually every human being with a shred of ambition. and they’ll all start waking up and we’re all going to be watching with popcorn. we’re still so early, chat. somehow.

@Gregpazo @ScriptedAlchemy @AdamRackis yeah very interested here 👀

@ScriptedAlchemy @AdamRackis Say more 👀

I recently tweeted something like "for frontend code I usually just accept what AI produces if the UI looks right" Boy was that stupid. My god what a stupid thing to say.

@emilkowalski Missed a trick her by not registering emilkowal.skill & linking it to that

Turned my blog articles into one big design engineering skill that you can use with coding agents like Claude Code or Codex. It covers animations, component design, principles from my open source projects like Sonner, and more. emilkowal.ski/skill

@rohanvarma Have a look at conductor.build and take inspiration from their UX (Claude only for now). How they handle informing the user about skill use, subagents etc is unparalleled

If you have tried the Codex App, but don't use it as a daily driver yet: What needs to improve to make it the primary place you work with coding agents?

@leerob @joshmatz @cursor_ai Is there anything in the works (even long term) to work across repos in the cloud? We have separate FE/API/CDK repos and it would be great to point and agent at all of them remotely to deliver a feature end to end.

@joshmatz @cursor_ai You get to pick the repo/environment to use. And you can set up your environment to have really anything installed in the VM!

We're introducing Cursor Automations to build always-on agents.

@flybayer @MarkLyck wait what. I didn't even know this was an option

@MarkLyck Our own REST api

I LOVE TANSTACK DB. The perf UX of our new dashboard is off the charts. TSDB makes it so easy to manage the data, prefetching, etc. ✨ Local first speed without local first tradeoffs ✨

Definition of locked in

@FaYiMoRa @schanuelmiller @thepanta82 Yeah same here for sure!

@schanuelmiller @thepanta82 Will definitely love to see docs on this. I assumed nitro was the only way too.

nitro is the most annoying part of tanstack start just give me a normal backend ffs

@Champdebloom @colinhacks @conductor_build Was gonna say that he was describing @conductor_build down to the buttons

@colinhacks Sounds a lot like @conductor_build, I only use Cursor for small “manual” changes these days.

fully local multi-tasking with Cursor is a mess. here's what I want: ✨ you open a new tab in the Agent pane ✨ it creates a new worktree AND checks out a new branch (name generated based on initial prompt) ✨ there's a persistent terminal at the bottom coupled to that task that you and the agent can use ✨ there are buttons to merge into `main` or create a PR ✨ all changes are isolated from all other agent tabs. ✨ changing tabs auto-checks out the associated worktree+branch

@imax153 @EffectTS_ I just implemented observability in a small app for work and getting effect logs into Datadog was a bit of a chore with de-trace. Had to convert them to Winston and use the Winston transport. Something native here would be awesome! Now that it is done it works like a dream!

Effect has always worked well with OpenTelemetry. In Effect 4.0, we’re making it even easier to connect your app to the observability tools you already use. For example, adding a Prometheus-scrapable endpoint to your HTTP server now takes just one line of code:

@kitlangton @aurelienqsw 100% I'm trying to effect pill people at work. I'd rather just make them watch this. It is much better than anything I would come up with

@aurelienqsw 🫶 Could you see yourself going through a (free) Effect course in this weird animated narrated slideshow format? 😅

Is this something?

@stephendahcrown @Prathamesh_7717 @tan_stack @honojs @golang @nextjs @tannerlinsley Yeah definitely. There’s a reason the world runs on Java backends

@cillo_b @Prathamesh_7717 @tan_stack @honojs @golang @nextjs @tannerlinsley Truth be told, most people have not built anything for public use. The more you explore the ecosystem and learn about good engineering, the less you make these weird comparisons. At this point, most new frameworks introduce too many issues to even be recommended.

I believe that having dedicated frontend(@tan_stack router) and backend(@honojs or @golang ) is better than having full stack framework(@tan_stack start or @nextjs ). what do you think @tannerlinsley ?

@stephendahcrown @Prathamesh_7717 @tan_stack @honojs @golang @nextjs @tannerlinsley 100% that’s why it annoys me when people push it as a replacement to new devs

@cillo_b @Prathamesh_7717 @tan_stack @honojs @golang @nextjs @tannerlinsley Yeah, exactly. I don't expect anyone to build the core of your business logic on NextJS. Even the simplest of applications.

@stephendahcrown @Prathamesh_7717 @tan_stack @honojs @golang @nextjs @tannerlinsley Well next would say that wouldn’t they. The reality is any serious application would benefit from a dedicated API built with a framework that is specifically built for backend development. Use Next or React Router or Start as BFFs

@Prathamesh_7717 @tan_stack @honojs @golang @nextjs @tannerlinsley I believe this ultimately depends on what you need. Most simple web applications would work well using this approach but you'd be giving up some form of upgradability. NextJs specifically recommends that the framework is not a replacement for backend

@RoccoVaccone @braelyn_ai Long running processes/async background tasks/independent scaling. Theres a tonne or reasons to use a purpose built backend framework

@braelyn_ai Fair point on separation of concerns. For complicated projects I use monorepos anyway for strong type safety across platforms. Why deploy another service instead of just separating the logic into a shared package?

Maturing is realizing that Next.js isn't it

@mixeddev @MichaelArnaldi @ericclemmons @tan_stack Yeah this would be a great thing to document

@MichaelArnaldi @ericclemmons Yes both are 🙌🏽 only thing is, the docs in RPC are missing info on how to plug 🔌 the server into the framework (@tan_stack start or next.js), even though the toWebHandler() function exists in the package. Had to explore the package source to figure out that’s what I was missing.

Torn between Hono RPC, oRPC & TanStack Start createServerFn, but I NEED STRONGLY TYPED ERRORS • ❌ Start doesn't support that github.com/TanStack/route… • 🤔 Hono has #parsing-a-response-with-type-safety-helper" target="_blank" rel="nofollow noopener">hono.dev/docs/guides/rp…, but DetailedError doesn't return actual types • ✅ oRPC has #type%E2%80%90safe-error-handling" target="_blank" rel="nofollow noopener">orpc.unnoq.com/docs/error-han…

@pierceboggan @wesbos @code What type of app do you build? Just at a high level? I’m thinking about implementing something similar but am not sure how much someone can actually get done in 30 minutes

More from the hiring manager perspective - We recently redid our interviews for @code product team. They all start with an app building exercise - you can use any AI developer tool you want. Super interesting, you learn a lot about how everyone uses AI tools differently. Some spend 30 minutes planning, then pass to AI and finish exercise in 5 minutes. Some iterate much more with agent. Some prefer to stay in editor. Then we have two interviews focused around getting to the root of developer problem and designing a solution to solve that problem (based on real feedback we've gotten on the team).

Who has been interviewing in the last ~3 months? What are things looking like? What does the interview process look like now?

@muddsicle @MichaelArnaldi @pavelsvitek_ @iamsahaj_xyz @EffectTS_ This is a good idea. I'm using effect for a new project at work, and context7 & the effect docs are helping, but an official source of truth would be good

@MichaelArnaldi @pavelsvitek_ @iamsahaj_xyz @EffectTS_ @MichaelArnaldi maybe y’all could publish an official AGENTS.md file? You’re right though, once you have enough effect code in a codebase, ai/cursor tab completion can autocomplete a good amount of the boilerplate. But starting from zero as a newcomer, it’s rough

@theo Imagine saying this when the mass effect trilogy exists

This is unironically the mindset within Bethesda. They will blame literally anything other than their dated game design.