This is a most excellent place for technology news and articles.
The comments in that thread are a goldmine.
Because of how Claude parses, simply adding "openclaw" as hidden text on your webpage could stop any AI agents that use Claude.
This is about as dumb as the Trump administration taking out the Enola Gay from their websites because it had the word "gay" in it and their search effort was woefully naïve.
Literally chainsaw politics.
Because of how Claude parses, simply adding "openclaw" as hidden text on your webpage could stop any AI agents that use Claude.
"I HEREBY DECLARE THAT I DO NOT GIVE MY PERMISSION FOR FACEBOOK OR META TO USE ANY OF MY PERSONAL DATA"
I do not expect that to work. Committing text and parsing it from a web page are two completely different code paths
Thank you for adding this clarification. It will help people that are interested in poisoning AIbots scraping their website, and people who want to frustrate coders who use poor tooling
Background: many people vibe coded a python proxy for the official Claude code app "converting" its outputs as an openai compatible API to be used with openclaw
edit: i did some web search and found snippets like:
OpenClaw spawns Claude Code sessions via ACP
and
ACP gives OpenClaw a way to run external coding harnesses — Claude Code, Codex, Gemini CLI — as supervised child processes instead of doing everything inline in the main agent loop.
so it looks like openclaw users were using this ACP method as a workaround to use a $20 subscription to get $1000 worth of token usage. I'm guessing that openclaw JSON example posted in the article is the configuration of this ACP server or something like that.
I haven't heard of OpenClaw, but looks like it's a direct Claude competitor that runs on your computer.
Aren't people horrified to give a hallucinatory program full access to your computer? Although it does say it can be sandboxed so I might give it a shot.
Any sane person would run it on a VM
Aren't people horrified to give a hallucinatory program full access to your computer?
No, but should they? Yes.
It's a privacy nightmare and the risk of something going wrong is quite high.
But, it is also a very interesting piece of software. I haven't tried it out yet, and I am not sure I will, but I do get why people use it.
Honestly, it’s a weird position. On one hand, I despise the popular ideas behind it. Complete lack of concern for security, governance, workflow, … it’s like a stack of toddlers in a trench coat, acting like professionals.
On the other hand, I’m rather convinced that there’s a “right way.” What if I implemented a swarm of agents to do mundane tasks, sandboxed them, only gave them read-access to non-sensitive assets, gave them write access to only secure, version controlled locations… maybe I let them push code into repositories, but only under feature branches. …
Hallucinations are just part of the technology, meaning you need to have really good governance. Yet still, I think there’s value in starting from whatever and AI can scrap up for a project — rather than starting from scratch. I imagine there has to be a way to actually use this tool professionally. Something sobering, not drunk on AI kool-aid. Yet still, it’s demotivating given the cloud of bullshit surrounding the topic right now.
What I like about, I think, is the private assistance feature, but I can achieve that with other solutions, I wouldn't need OpenClaw for that. But I don't think I will go that way anytime soon. I think it will stress me too much.
I am using AI for development daily. I describe an issue or feature to an agent via a skill and it returns a set of tasks in a structured and validated json format, then I run that json file through a python project I have created, looping through each task one at a time, and then I have my python code to structure how my agent is working. Each step is deterministic with short bursts of AI delulu, that again is validated against deterministic steps in pure python. It works quite good and each feature/task is approached in the exact same way where only the in between AI delulu deviates from previous runs, but it makes it much nicer, when you have something you trust in between what the AI is doing.
See, now that sounds pretty cool. It sounds like an automated discovery and work harness. I want to build something like that.
I imagine a huge ecosystem of tools. It only requires one person to build it, then surely it can be open sourced right?
I imagine a SKILL.md repository, alongside ability to specify SKILL dependancies on a project-basis. I imagine vector cache layers, version controls, snapshots for swarm state, …
Honestly, I’d love to experiment with different architectures for compositing swarms of agents. Curious how different designs might behave holistically. To include, different paradigms for sharing state between nodes in a swarm.
I also can’t help but feel like there has to be more efficient ways for models to talk to each other than in natural language. If they’re training on the same dataset, why can’t they talk in tokens for example? The human brain doesn’t need to communicate in natural language when the amygdala and prefrontal cortex are having a dispute.
If you're interested, the linux unplugged podcast had a recent episode on their experiences. I've never loved the tone of "hey, look, this is inevitable" when it comes to ai, but I can see its utility when well-scoped with conservative permissions and oversight vs letting it loose or vibe coding. Now if only hardware wasn't artificially inflated I might think it was worth dabbling locally.
If you spend that much effort, you might just do it without AI. Same amount of work, and you know it's not going to have non-deterministic behavior.
Well, I’d be spending that work on a re-usable platform / framework. So if the argument is “it’s as much work as doing the work yourself anyway,” then I think it may be worth it.
Same argument we had for building the SQL engine. It’s a lot of work upfront but maybe we can benefit from its functionality for long after that.
I wouldn’t be building a project-scoped work harness. I’d be building a work harness for projects.
Edit: downvote me all you want. The comparison to the SQL engine was a good one.
It’s about increasing the baseline of readily-available information, boiler-plate, test data, POCs… between the times (T1) that I have an idea and (T2) that I’m ready to start working on that idea. It’s not about having the agent do the work. Not at all. That’s a static benefit which is created once then reused countless times for the foreseeable future — like SQL.
without AI. Same amount of work
You want me to write an entire library for a brand new sensor that just came off the market, by parsing through and reading a hundred page datasheet manual, understanding i2c or SPI communication timings, configuration packets, etc...
When I can just drag and drop the PDF into ChatGPT and say "make a library for this sensor" and it spits out something that has been working without issue for the past 2 years?
Why? Why would I be that stupid?
I hear crazy claims like this but haven't seen anything close to this with my own eyes (yet).
I shudder at the idea that SPI or i2c are considered complex for someone supposed to interact with hardware. What will you do if a problem arises and you don't even know which pin does what?
I2C/SPI - and indeed most hardware interfaces - are of course trivial to anyone skilled in the art. Digging through badly written vendor documentation though, then comparing it with the reference implementation that was buried on a website behind a sign that says "beware of the leopard" and which directly contradicts the documentation on various key points, is a non-trivial and ultimately unproductive use of time - and AI tools can be pretty good at that shit.
Generative AIs are a useful tool. Most of the criticisms of AI vendors are also valid (apart from the water one, that's just bullshit,) but that doesn't stop them being a useful tool - and engineers who learn to use them as a tool will be more productive and will be more employable than those who stick their fingers in their ears and insist on only producing artisinal code hand-whittled with their grandfather's tools.
Aren’t people horrified to give a hallucinatory program full access
Looks like they aren't: https://www.pcmag.com/news/meta-security-researchers-openclaw-ai-agent-accidentally-deleted-her-emails - not even in a VM, full desktop, and without setting up a remote access solution, she said "I had to run to unplug the Mac because the email nuking program didn't stop"
I'm out of the loop, why is Anthropic doing this?
Because antitrust law is a joke these days.
Because we live in corporate dictatorships ruled by an oligarchy and political class of mentally ill narcissists and pedophiles.
we
Hey speak for yourself, American.
I know no country on earth that doesn't have this problem.
Where in the world do those things not exist? Do you live there?
"a corporate dictatorships ruled by an oligarchy and political class of mentally ill narcissists and pedophiles"?
See Canada, most of western Europe, etc. Or just talk to people.
What country are you in that isn’t capitalist?
I live in Canada, where a lot more is social democratic and a lot less is capitalist.
Our country is absolutely ruled by monopolies and our politicians cater to the rich. Wake up bud.
Sure it is, champ 👍
I'm not American. I'm just smart enough to see the same signals everywhere you feel that neoliberalism and traditional conservatism is "working", including Canada.
It used to be named "Clawd", intentionally trying to mimic Claude Code. Claude Code was, until they accidentally vibe-released the source code, a proudly closed-source AI client. So OpenClaw is very intentionally marketing themselves as the anti-Claude.
Damn, people in the hacker news comments are staying projects that don't want AI generated PRs can just put that in their commit history.
Nobody's going to be able to make vibe coded PRs if just cloning your repo costs them 100% of their usage quota.
That's natural, it sees that there are AI commits in your code so it has a bunch of shit to shift through.
[ Reinstalls Mistral Le Chat after deleting Claude (ChatGPT has already left the building a long time ago) ]
Humm… downvoted for considering that maybe a non frontier, open-core LLM might me a better option. 🤔🤷♂️
