Summarizer

For me the biggest benefit from using LLMs is that I feel way more motivated to try new tools because I don't have to worry about the initial setup. I'd previously encountered tools that seemed interesting, but as soon as I tried getting it to run I found myself going down an infinite debugging hole. With an LLM I can usually explain my system's constraints and the best models will give me a working setup from which I can begin iterating. The funny part is that most of these tools are usually AI related in some way, but getting a functional environment often felt impossible unless you had really modern hardware.

It might be role-specific. I'm a solutions engineer. A large portion of my time is spent making demos for customers. LLMs have been a game-changer for me, because not only can I spit out _more_ demos, but I can handle more edge cases in demos that people run into. E.g. for example, someone wrote in asking how to use our REST API with Python. I KNOW a common issue people run into is they forget to handle rate limits, but I also know more JavaScript than Python and have limited time, so before I'd write: ``` # NOTE: Make sure to handle the rate limit! This is just an example. See example.com/docs/javascript/rate-limit-example for a js example doing this. ``` Unsurprisingly, more than half of customers would just ignore the comment, forget to handle the rate limit, and then write in a few months later. With Claude, I just write "Create a customer demo in Python that handles rate limits. Use example.com/docs/javascript/rate-limit-example as a reference," and it gets me 95% of the way there. There are probably 100 other small examples like this where I had the "vibe" to know where the customer might trip over, but not the time to plug up all the little documentation example holes myself. Ideally, yes, hiring a full-time person to handle plugging up these holes would be great, but if you're resource constrained paying Anthropic for tokens is a much faster/cheaper solution in the short term.

Some ? I'd be shocked if it's less than 70% of everything AI-related in here. For example a lot of pro-OpenAI astroturfing really wanted you to know that 5.3 scored better than opus on terminal-bench 2.0 this week, and a lot of Anthropic astroturfing likes to claim that all your issues with it will simply go away as soon as you switch to a $200/month plan (like you can't try Opus in the cheaper one and realise it's definitely not 10x better).

You can try opus in the cheaper one if you enable extra usage, though.

And they are currently giving away $50 worth of extra usage if you subscribed to Pro before Feb 4.

> The C compiler that Anthropic created or whatever verb your want to use should prove that Claude is capable of doing reasonably complex level of making software. I don't doubt that an LLM would theoretically be capable of doing these sorts of things, nor did I intend to give off that sentiment, rather I was more evaluating if it was as practical as some people seem to be making the case for. For example, a C compiler is very impressive, but its clear from the blog post[0] that this required a massive amount of effort setting things up and constant monitoring and working around limitations of Claude Code and whatnot, not to mention $20,000. That doesn't seem at all practical, and I wonder if Nicholas Carlini (the author of the Anthropic post) would have had more success using Claude Code alongside his own abilities for significantly cheaper. While it might seem like moving the goalpost, I don't think it's the same thing to compare what I was saying with the fact that a multi billion dollar corporation whose entire business model relies on it can vibe code a C compiler with $20,000 worth of tokens. > The problem is people have egos, myself included. Not in the inflated sense, but in the "I built a thing a now the Internet is shitting on me and I feel bad" sense. Yes, this is actually a good point. I do feel like there's a self report bias at play here when it comes to this too. For example, someone might feel like they're more productive, but their output is roughly the same as what it was pre-LLM tooling. This is kind of where I'm at right now with this whole thing. > The "open secret" is that shipping stuff is hard. Who hasn't bought a domain name for a side project that didn't go anywhere. If there's anybody out there, raise your hand! So there's another filtering effect. My hand is definitely up here, shipping is very hard! I would also agree that it's an "open secret", especially given that "buying a domain name for a side project that never goes anywhere" is such a universal experience. I think both things can be true though. It can be true that these tools are definitely a step up from traditional IDE-style tooling, while also being true that they are not nearly as good as some would have you believe. I appreciate the insight, thanks for replying. [0]: https://www.anthropic.com/engineering/building-c-compiler

You must use the paid plans and get the pro / max subscriptions to get ultimate results The free versions are toys

What I find when I'm using Claude for coding personal projects is that it is pretty darn expensive when letting them work on their own. Is the cost of tokens ever a concern for those who use OpenClaw?

Mind you, that regardless of your sentiment towards OpenClaw, not everyone is able to afford a sparse Mac Mini (especially given ram prices) and a ton of Claude tokens/super beefy GPU for local models to run this stuff. That's to the supposed "democratisation of knowledge and technology".

FWIW Mac Minis have not increased in price because of "RAM Prices". Same models cost exactly the same as a year ago. Maybe it will change in the future, maybe not. Who knows. But right now Apple seems to have secure a good stash of RAM to use and avoid price changes.

I'm sorry dude but your last post was also hyping up R1 which was a total disaster. Do you mind actually sharing your experience with OpenClaw, such as how are you orchestrating a project? How much does it cost? How do you prompt it? What tasks do you get done? How much does it actually take to execute on those tasks? What is your interaction with the agent?

Since many posts mention lack of substance, providing a link to the All-In Podcast from last week in which they discuss Clawdbot (prior to re-brand). https://www.youtube.com/watch?v=gXY1kx7zlkk&t=2754s For the impatient, here's a transcript summary (from Gemini): The speaker describes creating a "virtual employee" (dubbed a "replicant") running on a local server with unrestricted, authenticated access to a real productivity stack—including Gmail, Notion, Slack, and WhatsApp. Tasked with podcast production, the agent autonomously researched guests, "vibe coded" its own custom CRM to manage data, sent email invitations, and maintained a work log on a shared calendar. The experiment highlights the agent's ability to build its own internal tools to solve problems and interact with humans via email and LinkedIn without being detected as AI. He ultimately concludes that for some roles, OpenClaw can do 90%+ of the work autonomously. Jason controversially mentions buying Macs to run Kimi 2.5 locally so they can save on costs. Others argue that hosting an open model on inference optimized hardware in the cloud is a better option, but doing so requires sharing potentially sensitive data.