Berthd | Hacker News Reader

Berthd

berthd.app

The more time goes on the more I think there is basically no moat in AI products because everyone is literally building the same shit. There is zero creativity in AI product right now. Yeah they might be different designs, but they are all the same exact product under the skin.

Cursor, Devin, Factory, Amp Code, Codex, Claude Codex, Antigravity, Conductor, OpenCode, T3 Code, Orca, Emdash, and hundreds of more of the same exact shit.

Yeah they might market it same fancy revolutionary technology like Orbs, but it not it just an agent harness local or remote. You can do orbs literally with Claude Code with a single prompt it not special.

We have AI that lets you build things freely and yet here we are all building the same exact thing.

This is just coding harness. The personal assistant space is the exact same.

what is Orbs?

I think it's like a Plumbus

Which btw is very similar to Lorpi

https://ampcode.com/what-are-orbs

Agents running in a remote vm.

weren't you all running agents in contained vms before???

Agreed. I think coding agents have shown us one very important thing: It was never about the code.

Your idea and how well you can execute on it, i.e. how well you understand the problem space, is what has always mattered.

Coding being so easy now has just pulled back the curtain and shown the world that software developers really don't have much left to contribute. It's been done, a thousandfold.

I feel like everyone already knew the code was the easy part, even pre-AI.

> Coding being so easy now has just pulled back the curtain and shown the world that software developers really don't have much left to contribute. It's been done, a thousandfold.

Maybe, but I feel like there's a reason that the "useful" AI-generated work all seems to come from people who could get to the same result given "infinite" time anyways.

AI tool usage reflects and amplifies your own skill sets IMO, and I think it's pretty hard to be amazing at software architecture design but be bad at coding.

> I feel like everyone already knew the code was the easy part, even pre-AI.

I don’t agree with this at all. There still tons of buggy software around, and we ship multi-gigabyte electron installs that make fanspeeds go high and are way less efficient than software written 20 years ago.

Also the largest problem of AI written code is that is so hard to maintain, so no, coding is not the easy part.

I think it is the other way round. Building requirements from many sources of data and view points is the easy part. Implementing great, maintainable systems out of them needs specialists.

I agree there's a bunch of buggy software, my contention is that a lot of that exists is because people are bad at the hard part and the easy part!

I am also dissatisfied at the current state of software so feel your pain. I think it's just that like .... I think if you can't get the code right you're probably not getting the requirements etc right either.

The velocity of how this space gets reskinned is wild.

Clearly we do not yet feel satisfied with the way we consume and work with agents at scale.

It looks like perhaps rather than skinning we need a more novel UX or workflow to emerge.

I can't help but think all of this is living on borrowed time and eventually it will look more like talking to your "assistant" that fields everything for you and maybe some new async interruption model that fits humans better.

Biggest bottleneck in my opinion still is how to solve for information density.

Thanks so much for saying this!

Even with a single instance of Claude Code, running I am producing so much code that it is impossible to even self-review.

Can we please focus on that?

We simply need neural implant technology to overclock our brains, making the world (and its Claude instances) appear to go slower.

You got downvoted but your joke made me think of the show The Dark Matter and how they had the concept of a hallway with infinite doors to represent the multiverse to the human brain that cannot comprehend the state of superposition.

I sort of get what you're going for maybe? In a sense these things are gonna evolve at a rate where we are not going to be able to operate at their plane and we will need a sort of UX translation

Self-review is an anti-pattern.

So much this. We don't review the assembly/machine-code that the compiler creates, why do we still review the code the AI generates?

If it works as desired then ship it; if it doesn't then iterate until it does.

That's such an AI bro take

Because the compiler does a deterministic transform on the input. And the input is less ambiguous than natural language and following more strict rules.

Why ask obvious things?

This is only OK where “it works” is the only criteria for accepting a software change.

Did you re-read your comment before hitting reply?

_because compiling a code language into machine language is deterministic_.

If you think that compiling and explaining in English something to an LLM so it outputs code is the same activity, I just hope to never have you as a coworker.

Don’t want to be an asshole but the pattern I have observed is that people who were not qualified / competent to code before November of 2022 have this mentality.

They want to make you feel dumb and useless with your software development knowledge.

If it works as desired then ship it

Before AI, this was a joke.

Proofreading an essay before submitting it is an anti pattern?

you don't proof read your speech writer. politicians don't even know what they will read until they start reading in the podium for example. ...not that i think it's good, but people have stopped caring mostly

And politicians are a famously well-loved group of people

The good ones are and I don’t even need to name names.

Good politicians do. They at least skim it to see where to add emphasis and the points they want to promote.

Really good politicians work with their speechwriters to get across exactly what they want to communicate.

Do you proofread the compiler’s output?

The compiler's output is reproducible and trustable in the extreme majority of the cases, as opposed to the piles of dogshit trained on the worst javascript practices that LLMs put out nondeterministically.

Just to be clear, when I say that it is impossible even to self review, it was an indication of how difficult is to review code of colleagues.

I built some tools that lets me view diffs and summon agents to execute changes. You can probably vibe code that in a few minutes.

This looks basically the same as what Paseo does, and Paseo is open source, works on lots of different platforms, and has native apps.

I have a homelab-type setup with macOS, phone, and ARM and x64 Linux platforms, all of which I can dispatch agent work to from my laptop from, like, an airplane.

It's extremely useful, scales up the number of experiments I can run, and decouples me from my dev laptop. But it's also just table stakes at this point.

I don't know if it is me but with Paseo I can't seems to resume a Claude code session. I feel I would prefer a mobile interface to herdr that is agent aware. Am I the only one not finding the right solution?

That's not a problem I've had, but I believe you. My thing here isn't so much "Paseo is amazing" (though: having something like Paseo is a pretty big power-up from directly working with model harnesses) as it is "whatever else you do needs to be meaningfully better than Paseo to be worth doing".

My real belief about this stuff is that these things are all light sabers: developers are just going to build their own.

How is this different from running Paseo [0] in your VM?

[0] https://paseo.sh/

Paseo seems to use a slightly less dark background color.

I am so tired.

The game is waiting out to see which jobs we can have instead.

I personally am very excited that the local hardware store is hiring.

All these agent conductors are like new frontend frameworks of the before times. Everyone is making and trying to market them.

Seriously. Who gets suckered into this hype? I find the default MacOS terminal to be plenty. Start a new tab, start claude, rinse & repeat.

Yeah they're just resume padding for AI engineers or something

These days, I run Claude Code on a headless VM and SSH in. I don't trust it to run unsupervised on my desktop. Maybe I'm paranoid.

Me. I don’t sit the whole day in front of the working computer or open laptop, but my phone connected to the always on server is always with me.

It seems like there's a similar project every week.

Yea. The people making these things are vibe coding them out the door before even doing the smallest little bit of market research. It's lunacy.

It is anxiety, not lunacy.

So why does Hacker News keep upvoting them?

sudo pmset disablesleep 1

Your Mac will now work while the lid is closed.

Exactly. My laptop just never goes to sleep, and it’s been great. What else do you need?

RIP your battery

A custom Mac application seems limiting compared to a web app. I use a web browser to connect from my laptop, tablet, and phone to use Linux VMs on exe.dev.

But it might be nice to have a nice way to run VMs on my Mac mini, too.

I don't get it. ChatGPT's Codex app allows you to use your machine, when its locked. It has full computer use, can code, open browsers, apps, etc. Why would we need a separate app for this?

They're milking "hourly cloud" charges which is absurdly high

It's so obvious on projects like this that everything's generated.

"One line. A worktree, a box, an agent." Fuck does that mean?

Have some pride and actually write the text yourself. Tell my why it's interesting. What problems are you trying to solve?

If you can't be bothered to write it, why should I be bothered to read it.

Why use brain when AI

Claude 5 hitting developers and product managers like 4o hit people who thought ChatGPT was their boyfriend or they were inventing a new branch of science.

Really just worth learning SSH and Tmux instead of building your workflow against something that's trying to replace them

https://rucksack.sh/

(similar)

Also created something similar. Love how people are working on similar things. This is what makes open source great and everyone has their own take on things.

"The agents run on your boxes, not on your laptop. Close it and they keep working; open it in the morning and what they did is there."

What is a box? Sorry is that a remote VM? A local VM that stays awake when the laptop screen is closed?

Was wondering the same. So I guess a box is a linux host? So this is ssh + tmux or zmx or screen?

Frankly 'box' feels like classic Claude-slop of 'throw a random synonym in that is less precise/internal jargon'. I wish Claude would fix this (if my guess is correct and the marketing website was generated and is claudeslop).

Projects like this need to start supporting coop et al.

So many apps do the same thing somehow. We need something like Zed's Delta!? Right? All Ai provider. Self hosted. Open source. As simple as possible. Right?

Isn't this built in to claude anyway?

How does it compares to native claude rc?

On the claude side, it periodically signs me out, breaking the session. Yes, I can ssh and reconnect but why? You can use a long-lived token but you are not allowed to run remote control for those sessions. Anything that solves that pesky problem is better than “Claude rc”.

How does this have so many upvotes?

Because anyone who says you can't buy GitHub stars or HN up votes is lying.

How can I sell github stars? Is there a marketplace for that? My HN upvotes are for sale as well.

I would trade my HN account for a larger SD, or a good memory stick.

Orca clone?

This thread is way too negative. I, for one, am so excited about all the iteration in this space. So many incredible, open source tools are being built - and Berthd looks like a great addition.

What an incredible gift to build this as an open source project. Thank you and I'm excited to give it a try!

Especially with half of the thread being unable to read or see the GitHub icon implying it’s a marketed product when it’s self hosted. Even if the docs are insanely slopped: if they’re to be believed and it works it is well beyond any locked in cloud runs/orchestration from Anthropic or OpenAI.

So it's another codex/claude cloud or modal?

What's this?

https://learn.chatgpt.com/docs/cloud

Yet another Vibe coded slop clone of a remote claude code.

The quality bar is quite low these days.

I don't know why this is getting downvoted. It's completely obviously that this is a 100% vibe coded slop project which is already supported with out of the box tools. There was, and is, almost zero need for this.

Reply on news.ycombinator.com