Clawdbot

Absolutely depressing. The unconditional delegation of even basic tasks to this thing. The dependence on it that will inevitably take place. Also, think about the level of access to your stuff this thing is allowed to have. What happens when a bad actor takes advantage of that? How secure is this thing?

The lack of security is the big issue for me. I don’t trust these companies to have that kind of access. Especially not together, connected in some way

"What happens when a bad actor takes advantage of that?"

You don't even need a bad actor, just an accidental, incorrect action from the LLM. It's funny to me that I'm running opencode in Docker just so it can't fuck with unversioned files, and other people are like, "Here's my whole life, go wild!"

They all sounds so giddy, like small children with their new toy, until the disaster happens.
Are these serious people or just bloggers who pretend to know software, and just write for clicks.

Gotta clear out my RSS/Instapaper backlog to make room for the inevitable trove of stories about people growing dependent on these services, only to watch them go belly up when there’s no money to be made. Bricked smart appliances have nothing on what’s about to happen next.

@Vajk The main developer, Peter Steinberger, is a respected developer who built the best iOS/Android library for working with PDF files.

I do find it all rather scary, and I hope to see better ways to isolate what it can do, but I think this is something to watch.

> They all sounds so giddy, like small children with their new toy, until the disaster happens.

Conspiracy theory - are these people just peddling this shit so nobody else takes the time to learn to code to make themselves more valuable? Tell the kids that coding is an obsolete skill so nobody else learns it.

I'm still skeptical that writing essays to an LLM and then having to spend time reading the code it generates to make sure it's good is the "new interface." Of course if you don't care and you don't read the code - I guess you're more "productive." While I think LLMs can be useful I think some people are getting a little carried away here.

@Michael Tsai
I am not keen to let an LLM wrapper, written largely by LLMs(and bragging about it), loose on anything.
No one who cares about safety, privacy and security should be ok with that.

I’m pretty skeptical of Peter’s decision-making process. He’s brilliant but unchecked, kind of like Tony Stark. Right now, his pinned tweet from December 29, 2025, says: “Confession: I ship code I never read.”

That said, he definitely has his fans and plenty of contributors. I’m intrigued by the integrations, especially exactly how they automated Messages.app. If I remember correctly and nothing has changed, you used to have to quit or relaunch Messages.app just to get it to write its data to disk so you could read new incoming messages.

Seeing as the whole project could be useful running on a separate machine in a user account with its own iCloud credentials, I’m also planning to check it out to see how it fits my privacy and security standards.

"I'm still skeptical that writing essays to an LLM and then having to spend time reading the code it generates"

That's the trick. They don't. They just hope that issues eventually fix themselves.

Maybe they're right, and LLMs get better faster than they can generate shitty code, so as their cognitive debt grows, so does the LLM's ability to do their tasks. Slowly, the human is replacing itself with the machine.

Or perhaps they're wrong, and LLM progress slows down, and they end up with tens or hundreds of thousands of lines of code, no idea how anything works, and no ability to make progress or fix bugs.

We'll see.

> He’s brilliant but unchecked, kind of like Tony Stark. Right now, his pinned tweet from December 29, 2025, says: “Confession: I ship code I never read.”

Maybe he's brilliant, maybe he's not but it really doesn't matter the only thing that matters is what he's doing. At the very least I think he's putting his reputation on the line. People should be aware that he's shipping code that he isn't reading. We all put trust in code we don't read to some extent (for ex: using system frameworks etc.) but code not even written or evaluated by a human well that still seems crazy to me - it's a good thing as of right now this vibe coding craze is relegated to dinky consumer apps.

More importantly what does this teach the next generation when we have developers who know how to code but don't want to basically proclaim the skill of coding is obsolete? You're encouraging new people not to learn anything. So the next generation of "app makers" how are they going to know how to solve a bug? Change the model and just hope it works. Maybe he can work his way out of a hairy situation because he has dev experience but what about the fucking children?

> Maybe they're right, and LLMs get better faster than they can generate shitty code, so as their cognitive debt grows, so does the LLM's ability to do their tasks. Slowly, the human is replacing itself with the machine.

I think they are wrong. My current approach--write code, read code, use an LLM as a tool when it makes sense still seems like the right choice right now. If I'm wrong in the future...well... the idea that this is some "hard skill" so you better start talking to your LLM and stop wasting time reading code so you don't become obsolete is funny because in that world you are mostly obsolete anyway (and presumably this would apply to other professions). If you're smart enough to learn how to write an app the old fashioned way - getting "up to speed" in the future world where you ask an LLM to do everything for you should be a breeze.

What is even happening right now in the AI pond?

At the beginning of the year Steve Yegge came out with Gas Town which he created his own terminology to describe things that already have a name in his LLM driven workflow...thing. I would have chalked it up to one of his bombastic posts, but it even came complete with a supporting crypto rug pull.

Now we have clawdbot where people praising it sound somewhat unhinged.

I think LLMs can be helpful, if used judiciously, but all of this just adds to the feeling that something very bad is about to happen with all of this AI hype.

sudo adb install malware

I get the caution, but there's a lack of perspective here.

Yes, trust is a real issue, but last time I checked, it was for humans too.

Anyone who employs more than one coder likely ships code they haven't read.

"Anyone who employs more than one coder likely ships code they haven't read."

But a human did.