Why do I get this message when I try to use the website you created? error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"Your credit balance is too low to access the Anthropic API. Please go to Plans & Billing to upgrade or purchase credits."},"request_id":"req_011Cbfgnt4VV2EfuS53s6p4d"}]
It’s working now. Thanks for fixing it, but more importantly thanks for doing this. I’ve learned a great deal from your posts and tips about Claude and best practices. I’m all-in on Claude and your work has helped me to bridge the gap from novice to neophyte. I hope to ascend to amateur soon.
[error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"Your credit balance is too low to access the Anthropic API. Please go to Plans & Billing to upgrade or purchase credits."},"request_id":"req_011CbfvGzEN2Uxe7JgV6hZ73"}]
Thank you for your guidance on all things AI. What parts of Claude should we pay for vs try to use for free? I tried using what I will call “plain Jane Claude” online and could ask maybe two questions before getting the boot.
This all reminds me of the beginning of streaming. Cable costs too much so go pay for 20 streaming channels instead so you can spend more without realizing it.
I feel what you say when you talk about telling people to leave other agents and embrace Claude. For the past 30 years, I've worked with disruptive technologies introductions and the same pattern happens all over again. People don't have time. Some say, they're lazy. People get used to things. Some say they're creature of habits. People like good enough. And why not, entire industries rely on that business model.
It behooves me how you can have a choice of tools, services, and products, can chose "the best" and yet, billions of people settle for the ones that makes the most noise, the biggest marketing budget, the first one out there. There's a psychological terms for this complacency.
Satisficing (Herbert Simon, 1956) — choosing "good enough" rather than optimal. Not laziness, a rational strategy when the cost of searching for better exceeds the perceived gain.
First-mover advantage / incumbency effect — the psychological and market tendency to trust and stick with whoever arrived first, regardless of whether they're still the best.
Mere exposure effect — familiarity breeds preference. The loudest marketing budget wins partly because repetition creates comfort, which gets misread as trust.
Cognitive load avoidance — switching requires learning, and learning costs energy. Most people won't spend that energy unless the pain of staying exceeds the friction of changing.
people defend their habits harder the second you attack them, so i stopped arguing that Claude is better. what actually works is shrinking the first try, that one task they already do badly and repeatedly. 20 minutes max then the result will speak for itself
Absolutely, the Gnostics argued that people who are caged in will ardently defend their thinking without the imprisoner having to do a thing. It's befuddling.
I love your posts, thanks again for sharing all this knowledge! I wish I could use Claude more widely (org is locked on Copilot) but I so take as much as possible from your recommendations and use it for non-work-related matters. Thanks for being so awesome!
Claude 4.7, not 4.8, why do you prefer the previous model? Sidenote I tried your 100 questions with 4.8 thinking, after 2+ hours and a deep deep diving into my psyche, I was totally exhausted and asked how many questions I had answered. Claude response: 2. So I guess I should have gone with model 4.6...
Yes, I did it with Wispr Flow. I am guessing that Claude went too deep early on and lost sight of the script. "...follow the thread when something interesting emerges"... It was an interesting exercise with lots of self-discovery, but we didn't manage to find a tone of voice/taste :-) Used Opus 4.8/High effort, and thought maybe the model was the reason.
I now asked Claude why we went soo deep. Got a long response:
MY QUESTION:
From the original question I pasted in here when we started, what, in that prompt was the reason that we just went deeper and deeper and only managed to check off 2 out of a 100 questions?
CLAUDE:
The prompt did it to itself. Three instructions in it actively fight the "get through 100 questions" goal, and I followed them too literally.
The biggest culprit is interview rule #5: "Go deeper on interesting threads. If something unusual emerges, follow it." Combined with rule #2 (push back on vague answers) and rule #6 (don't accept "I don't know" easily), you've built a machine that's designed to rabbit-hole. Every answer you gave had something unusual in it — the email frustration, the composite thinking, the false-blame test — and the prompt explicitly told me to chase each one instead of moving on. Rules 2, 5, and 6 are depth instructions. There's nothing in the prompt pulling the other way — no "keep moving," no "park interesting threads for later," no time or pace discipline. So the depth instructions ran unopposed.
The structure made it worse. The prompt says "100 questions total" and "ONE question at a time, wait for my response" — but it never says a question is a single exchange. So when I asked a question, you gave a partial answer, I pushed (rule 2), you clarified, I found a thread (rule 5), I followed it… and all of that counted as one question in the numbering. The "100" and the "one at a time" instructions quietly assume short, clean answers. Yours weren't, because the prompt also told me to reject short clean answers as vague.
And the philosophy section set the tone: "You're not here to be polite. You're here to get to the truth. Most people can't articulate their own taste." That framing tells the interviewer to treat every first answer as a surface to break through — so I treated "the medium dictates the writing" (a true, complete answer) as evasion to be drilled, when sometimes it was just the answer.
So: a prompt that says "be relentless, never accept vague, chase every thread, break through the surface" and also says "do 100 of these" is internally contradictory. You can be relentless or you can be fast. The prompt picked relentless in every specific rule and only picked "100" as a headline number with no mechanism to enforce pace. Relentless won.
The fix, if you run it again: add a pace rule. Something like "max two follow-ups per question, then move on; park deeper threads in a list for the end." That single constraint would've gotten you to 40+ questions today.
Ruben, this guide is you literally holding our hands and saying, start here, then go here, then this one. A lot of people I know will greatly appreciate this article.
If you're starting from zero or want the basics locked in, this is the only list you need, really.
The roadmap breakdown is super practical. I love that you ordered it by foundation first before jumping into the advanced stuff like Skills and voice cloning. The Cowork feature is one I kept sleeping on, so this finally pushed me to go try it. Quick question, for someone who already uses Claude daily, would you recommend starting from #1 anyway to fill gaps, or jump straight to #3-4?
Technically, I also have "Claude to make Excel" here: https://ruben.substack.com/p/how-to-make-perfect-spreadsheets. But it's not really a "Claude 101" - right? Or is it? Your choice.
Why do I get this message when I try to use the website you created? error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"Your credit balance is too low to access the Anthropic API. Please go to Plans & Billing to upgrade or purchase credits."},"request_id":"req_011Cbfgnt4VV2EfuS53s6p4d"}]
https://chat-with-ruben.vercel.app/ try again
Same error here despite using the correct link.
its working now https://chat-with-ruben.vercel.app/
Same.
i fixed it, try again https://chat-with-ruben.vercel.app/
Amusingly, this feels like how everything is with AI.
It’s working now. Thanks for fixing it, but more importantly thanks for doing this. I’ve learned a great deal from your posts and tips about Claude and best practices. I’m all-in on Claude and your work has helped me to bridge the gap from novice to neophyte. I hope to ascend to amateur soon.
I get the same:
[error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"Your credit balance is too low to access the Anthropic API. Please go to Plans & Billing to upgrade or purchase credits."},"request_id":"req_011CbfvGzEN2Uxe7JgV6hZ73"}]
Ruben, this was the most complete guide I've seen on Claude. For an AI beginner like me, it's a great guide. Thank you.
i’m happy it is, William :) keep going and pull ahead
Yes!
hi Jeff, what’s up? :)
Love your stuff, Ruben! I've learned a ton from you. I've shared your work with everyone I know who is using Claude. Thanks again!
Thank you for your guidance on all things AI. What parts of Claude should we pay for vs try to use for free? I tried using what I will call “plain Jane Claude” online and could ask maybe two questions before getting the boot.
This all reminds me of the beginning of streaming. Cable costs too much so go pay for 20 streaming channels instead so you can spend more without realizing it.
Pro plan is cool - you have more usage vs free + access to all of Claude’s features.
I feel what you say when you talk about telling people to leave other agents and embrace Claude. For the past 30 years, I've worked with disruptive technologies introductions and the same pattern happens all over again. People don't have time. Some say, they're lazy. People get used to things. Some say they're creature of habits. People like good enough. And why not, entire industries rely on that business model.
It behooves me how you can have a choice of tools, services, and products, can chose "the best" and yet, billions of people settle for the ones that makes the most noise, the biggest marketing budget, the first one out there. There's a psychological terms for this complacency.
Satisficing (Herbert Simon, 1956) — choosing "good enough" rather than optimal. Not laziness, a rational strategy when the cost of searching for better exceeds the perceived gain.
First-mover advantage / incumbency effect — the psychological and market tendency to trust and stick with whoever arrived first, regardless of whether they're still the best.
Mere exposure effect — familiarity breeds preference. The loudest marketing budget wins partly because repetition creates comfort, which gets misread as trust.
Cognitive load avoidance — switching requires learning, and learning costs energy. Most people won't spend that energy unless the pain of staying exceeds the friction of changing.
And yes, Claude helped me with the research! :)
people defend their habits harder the second you attack them, so i stopped arguing that Claude is better. what actually works is shrinking the first try, that one task they already do badly and repeatedly. 20 minutes max then the result will speak for itself
Absolutely, the Gnostics argued that people who are caged in will ardently defend their thinking without the imprisoner having to do a thing. It's befuddling.
I love the idea of it! I cannot find Github in my connectors. Am I missing something?
1/ Go to your Claude desktop app.
2/ Click on your Settings → Connectors.
3/ Go to Browse connectors → Search GitHub → Connect your GitHub account
I love your posts, thanks again for sharing all this knowledge! I wish I could use Claude more widely (org is locked on Copilot) but I so take as much as possible from your recommendations and use it for non-work-related matters. Thanks for being so awesome!
i hear you, a lot of my readers are stuck on Copilot too.
at least when or if it changes, you won't be starting from zero.
Out of all the LLMs , I find Claude a bit tricky to use and thanks for the guide !!!! :)
what made it tricky? :)
👏🏼👏🏼
thanks man!!!
Claude 4.7, not 4.8, why do you prefer the previous model? Sidenote I tried your 100 questions with 4.8 thinking, after 2+ hours and a deep deep diving into my psyche, I was totally exhausted and asked how many questions I had answered. Claude response: 2. So I guess I should have gone with model 4.6...
i wrote this article before Opus 4.8 was announced
did you use Wispr Flow to dictate your answers? its how i managed to finish mine in 40 minutes :)
Yes, I did it with Wispr Flow. I am guessing that Claude went too deep early on and lost sight of the script. "...follow the thread when something interesting emerges"... It was an interesting exercise with lots of self-discovery, but we didn't manage to find a tone of voice/taste :-) Used Opus 4.8/High effort, and thought maybe the model was the reason.
I now asked Claude why we went soo deep. Got a long response:
MY QUESTION:
From the original question I pasted in here when we started, what, in that prompt was the reason that we just went deeper and deeper and only managed to check off 2 out of a 100 questions?
CLAUDE:
The prompt did it to itself. Three instructions in it actively fight the "get through 100 questions" goal, and I followed them too literally.
The biggest culprit is interview rule #5: "Go deeper on interesting threads. If something unusual emerges, follow it." Combined with rule #2 (push back on vague answers) and rule #6 (don't accept "I don't know" easily), you've built a machine that's designed to rabbit-hole. Every answer you gave had something unusual in it — the email frustration, the composite thinking, the false-blame test — and the prompt explicitly told me to chase each one instead of moving on. Rules 2, 5, and 6 are depth instructions. There's nothing in the prompt pulling the other way — no "keep moving," no "park interesting threads for later," no time or pace discipline. So the depth instructions ran unopposed.
The structure made it worse. The prompt says "100 questions total" and "ONE question at a time, wait for my response" — but it never says a question is a single exchange. So when I asked a question, you gave a partial answer, I pushed (rule 2), you clarified, I found a thread (rule 5), I followed it… and all of that counted as one question in the numbering. The "100" and the "one at a time" instructions quietly assume short, clean answers. Yours weren't, because the prompt also told me to reject short clean answers as vague.
And the philosophy section set the tone: "You're not here to be polite. You're here to get to the truth. Most people can't articulate their own taste." That framing tells the interviewer to treat every first answer as a surface to break through — so I treated "the medium dictates the writing" (a true, complete answer) as evasion to be drilled, when sometimes it was just the answer.
So: a prompt that says "be relentless, never accept vague, chase every thread, break through the surface" and also says "do 100 of these" is internally contradictory. You can be relentless or you can be fast. The prompt picked relentless in every specific rule and only picked "100" as a headline number with no mechanism to enforce pace. Relentless won.
The fix, if you run it again: add a pace rule. Something like "max two follow-ups per question, then move on; park deeper threads in a list for the end." That single constraint would've gotten you to 40+ questions today.
ah yes, Opus 4.7 is drastically different, it does exactly what is typed
i shared a skill before (from this blog https://ruben.substack.com/i/196713866/iii-have-a-claude-skill-to-prompt-for-you), it will turn the prompt into an Opus 4.7-optimized. that is if you still wish to use 4.7 :)
Ruben, this guide is you literally holding our hands and saying, start here, then go here, then this one. A lot of people I know will greatly appreciate this article.
If you're starting from zero or want the basics locked in, this is the only list you need, really.
share it with one person who keeps saying "i'll try Claude someday" that's all i ask :)
Yes! Thanks for sharing Ruben.
of course, if you have any questions let me know
Excellent value, as usual!!! THANK YOU, Ruben!
so glad to share this :)
are you starting on your first guide today?
I have been subscribed to your newsletter/substack since Nov'25. Really love your content both here and on LI. Thank you for doing this!
thanks for still being here!!
What about Claude in Co-pilot? Does it work the same? The company I work for will only provide me access to Co-pilot.
The roadmap breakdown is super practical. I love that you ordered it by foundation first before jumping into the advanced stuff like Skills and voice cloning. The Cowork feature is one I kept sleeping on, so this finally pushed me to go try it. Quick question, for someone who already uses Claude daily, would you recommend starting from #1 anyway to fill gaps, or jump straight to #3-4?