Ok I got Stacks and MKStacks working and I've started vibe coding a LCARS nostr client. A few crashes and hiccups here and there but I'm making progress. It is pretty fun to type a couple sentences and then beep boop here's a website.

I've already burned through about 14,000 sats worth of sonnet tokens through nostr:nprofile1qqsdy27dk8f9qk7qvrm94pkdtus9xtk970jpcp4w48k6cw0khfm06mspp4mhxue69uhkummn9ekx7mqpz4mhxue69uhkummnw3ezummcw3ezuer9wchsj08ren and at this rate I feel like I'd probably use 50,000 more sats before getting a decent product. Maybe thats a good deal for a non-coder to bring a custom nostr client to life? I'm just curious to hear others experiences regarding AI token costs to build things. Does that seem reasonable, more or less? nostr:nprofile1qqsfm7l5mygtkeecks5d5qzq4tg9tmlnszxhzxqxqhtf4c8gy5rldyqpr9mhxue69uhhyetvv9ujumt0d4hhxarj9ecxjmnt9upp55xs nostr:nprofile1qqsykxqyeyx4nhl3jhhqarmf9wv203mzhute8vlpymz5d4esmjcygacppemhxue69uhkummn9ekx7mp0qyvhwumn8ghj7un9d3shjtnndehhyapwwdhkx6tpdshslfjd7kOBV nostr:nprofile1qqsr7acdvhf6we9fch94qwhpy0nza36e3tgrtkpku25ppuu80f69kfqppemhxue69uhkummn9ekx7mp0qy0hwumn8ghj7mn0wd68yttjv4kxz7fwv3jhyettwfhhxuewd4jj7lal3ny nostr:nprofile1qqsfev65tsmfgrv69mux65x4c7504wgrzrxgnrzrgj70cnyz9l68hjspz4mhxue69uhkzem8wghxummnw3ezumrpdejqzxthwden5te0dehk7unwdajx2tnwdaehgu339e3k7mgexwks6 nostr:nprofile1qqst03hkj9w04xnzllm2rupxqn0g3s3ud3kx6xu0vtrucyr5nuc8aqgpzamhxue69uhk2cmvd9c8xefwwp6kytmjv4kxz7gpzemhxue69uhk2er9dchxummnw3ezumrpdejz7mx5k3k

Reply to this note

Please Login to reply.

Discussion

MKStack, excuse me

Of course it’s worth it from a value for money perspective.

If you want to build an app like the ones MKstacks is producing it will probably cost you in the millions of sats to pay a dev and many hours.

No doubt and I am grateful that it exists and am having fun playing around. But I don't do this for a living and could easily see myself spending hundreds of dollars before I know it on something that was just a fun idea. So I'm just trying to gauge what a reasonable amount of money to play around with is so I don't kill my own vibes.

It's all about the prompting. Sometimes I burn through tokens on bugs. The key is to not try more than two times on the same bug or reprompt. I've found it's best to roll back and reprompt. But in the grand scheme of things, this is cheaper than paying a human developer to do this. And faster. And the process is probably easier too.

Great thanks for your perspective. Will keep in mind re: fixing bugs. I am not a developer or someone with lots of extra disposable income so the beauty of the whole thing to me is that I can try to bring an idea to life for fun that I would never even consider or know where to begin otherwise.

In my experience it's about $10 USD to get the first MKStack prompt completed with Claude Sonnet 4. After that, it depends a lot on what you are trying to do. For a super simple app you could probably get an MVP in ~$100. For a more detailed client that does a lot of things, easily you could spend $2,000+ on Claude.

It's a balancing act. Because, on one hand, how much would it cost to hire a human to do it? Or, how much is your own time worth to you?

You do start to get a sense of what tasks are worth vibing and which are not. Ex. If I know it would take me less than a couple minutes to just do it myself, I ususally do. You can also consider swapping agents for tasks that are simpler (stacks agent -m gemini-2.5-pro-preview). However, IMO Gemini is pretty stupid with code compared to Claude - so even if its cheaper if it takes more than two tries it might do more damage than it's worth.

Good prompting also helps reduce the amount you spend on fixing errors or miscommunications - and that takes practice.

Just my experience! Depends heavily on your final goal :)

Thanks for the thoughtful response it is very helpful

Happy to help! ☺️

Great response thank you

I'm curious if the prompt caching was working properly when you were using MKStacks. I had another user report that it wasn't. If that's the case, you would have been paying about double what it should have been.

If you DM me your credit id I can look it up and see if the prompt caching was happening or not.

But for some reason Goose and this new Stacks tool have trouble prompt caching. I'm not sure if it is PPQ's fault or the tools' fault, but I've been investigating.

nostr:nprofile1qyg8wumn8ghj7mn0wd68ytnddakj7qgkwaehxw309ahx7um5wfjkc6t5v4ejummjvuhsqgzqh8y9lll2lsw2m7xrpf89ezrxplmwf9c6phrj84dtva94ucd52yr8jvw7 also built a tool which intercepts the prompts from these tools for deeper examination. It may play a role in diagnosing the issue.

https://github.com/gzuuus/transparent-oai-api