Max Chepurin

Claude Code Claude Cowork Claude Design Claude Finance Claude for Teachers Claude for students Claude for babies Claude for eating Claude for drinking Claude for speaking Claude for hanging out with your friends Claude for talking to girls Claude Fucking Claude Marriage Claude for raising your child Claude for going and telling your neighbor to turn the fckn music off at 2AM Claude for coffee Claude for vacation Claude for enjoing your life instead of you Claude for shitposting on X Claude for doom scrolling Claude Dreaming Claude Breathing Claude Shitting Claude Laughing And on top of that MCP for plugging it into your ass

@ochuwa22 @NanouuSymeon Can relate 100%

@NanouuSymeon Watching movies. So low effort. I just want to rest without racking my brain

As a Developer, what is your hobbie after work?

ok, now commit it but remove "Co-Authored ..." part

@TRPage_dev My mom at the stage where she reads and trusts Gemini overviews whenever she googles smth. But Im already impressed! She now thinks she is AI power user😎

We’ve reached whatever stage of AI it is where I’m pretty sure my not very tech-savvy mother has been using AI in her text message responses…

I feel like Im in an abusive relationship at this point

I keep seeing this shit in agent prompts /instructions: "be thorough" "write clean code" "make it high quality" “make no mistakes😭”(this one is joke mostly but still) How do you guys expect it to work? Lets explore a scenario: - You are really good programmer - You land a job at new company - Instead of an onboarding process, documentation, or any guidance, all you get is: “Hey buddy, just dont make mistakes and itll be fine, alright?” Would that make you feel confident about the work you’re about to do?

@DeepStarts AI already writes better code than 95% developers

Software Developers, what’s your backup plan if AI writes better code than you in 3 years?

Im wondering what is the reason mentioning something like “Im pretty tech savvy”, “making wesites for 2 decades” in every corner. Who asked bro?😭 Does it have to do something with using apps, which is like basic knowledge? For me, it’s like going to buy a cheeseburger and saying something like: “Just so you know, guys, I’m a big tech-savvy boy.” I mean yeah whatever, man. Thats $6.30, please. Now go do whatever tech-savvy things you can imagine with this cheese.

@faridmovsumov Yea I guess so, I was wondering whether I supposed to fight on each and every user or not. It’s pretty funny because he first reached out in the cs with that

Brooo 😭😭😭 Should I even try to get him at this point? Do I even want to deal with a user like this? Should I actually take this feedback seriously?

Im gonna add X to my Linkedin “Post reader”

Is that even legal?😳

Unpopular opinion: Vibe coding isn’t as simple as everyone thinks.

Identity verification for Shopify Partners. Thoughts. Collaborator access is one of those Shopify flows where the merchant is basically saying: "yes, random developer from the internet, please enter my live store" So I kinda understand why Shopify is starting there. (Even though it sounds like another reminder of the direction the internet is heading. Im not the biggest fan of the broader trend toward a more regulated internet, but in Shopify's case this change actually seems reasonable.) I’ve also heard stories about people pretending to be Shopify agencies with lookalike names and emails, reaching out to their old clients, getting access to a store, taking the payment, and then disappearing. For legit partners this is probably annoying. For merchants, it may remove at least some of the obvious scammy behavior around collaborator access.

@TheStormDev I hope it wont be integrated into everything, they all ready consuming so much resources from Mother Earth, if there is a limit somewhere its resources

@maximchepurin Inference will be intrgrated into everything, its not about chat bots or agents. I think we cant imagine all usecases yet but there will be lots and lots of things with integrated intelligence. For the providers its a race to the bottom imho as it will be a commodity

You have to be unemployed to keep up with all of this. If Im not mistaken, this is already the third model release this week, and the fifth or sixth this month. Actually, forget what I just said. Im not even sure unemployed people could keep up with this shit. Im starting to think there might actually be a bubble. I just cant believe there is enough demand for all these tools.

But is everyone really using these models? My mom uses Gemini, but only because Google generates those AI summaries😂 These new models are supposed to compete primarily on coding. But are there really that many developers in the world to satisfy the demand? It's not like you can build data centers full of GPUs just for fun. They need to be generating inference, and people need to be paying for subscriptions. Otherwise its not profitable. Damn even those AI giants not that profitable as they should be rn. To make an analogy with electricity, imagine there were 20 electricity companies, all claiming their electricity was somehow better. I bet 99% users simply wouldn't care. All they care about is getting electricity itself. If one coding model is 2% better than another and I miss that release, that's fine. I can live with my coding agent finishing the task in 2min 35sec instead of 2min 4sec, or producing code thats 3% sloppier than another model's.

@maximchepurin Its a race definitely, but tokens are a new basic resource same as electricity.

@jzazove @iliveoffgrid

As promised, Shopify App Pricing just leveled up today: • More plans: 8 public / 15 private • No-charge plan testing in review + dev stores • App Events now accepts negative + fractional values shopify.dev/changelog/app-…

Shopify’s new Collections API is starting to roll out to development stores - and it changes how collections are created. Instead of choosing between manual and smart collections, every collection is now built from one or more sources. There are four source types: 1. Products You can include or exclude products using conditions such as product attributes and metafields, while also selecting individual products manually. It gives merchants more control, but it is more of an evolution of smart collections than a completely new feature. 2. Variants This lets you include only specific variants of a product in a collection. However, those variants do not become separate product cards. The storefront still shows one product card, with only the included variants available in its swatches or options. That may be useful for narrowing a collection, but it does not solve the common use case of displaying each color or variant as an individual product card. It can also feel confusing when the collection card shows fewer variants than the product page. I truly don't understand the reason behind it. 3. Collections A collection can now use other collections as its source. For example, an “All Snowboards” collection could automatically include products from “Men’s Snowboards” and “Women’s Snowboards.” People are calling these sub-collections, but that is not entirely accurate. Shopify is not introducing nested collection URLs, new Liquid relationships, automatic breadcrumbs, or a true collection hierarchy. It simply combines products from multiple collections into another collection. 4. Apps This is probably the most interesting source. Apps can provide their own logic for deciding which products belong in a collection. A reviews app, for example, could create a source containing the highest-rated products. A merchandising app could build collections based on external data, inventory signals, customer behavior, or custom business rules. Overall, the new API gives merchants and app developers much more flexibility. But variant collections are not separate variant cards, and collection sources are not yet true sub-collections. I’m currently building an app on top of the new Collections API and will share what I come up with next.

I just discovered that we have an Advisor tool in Claude. When to use it. The advisor fits these configurations: You currently use Sonnet on complex tasks: Add Opus as the advisor for a quality lift at similar or lower total cost. You currently use Haiku and want a step up in intelligence: Add Opus as the advisor. Expect higher cost than Haiku alone, but lower than switching the executor to a larger model. Results are task-dependent. Evaluate on your own workload. So many updates are happening every day that it’s easy to miss the majority of them. As far as I understand, this tool was released in beta back in March.

I keep seeing people compare AI models by cost per token. Which one is more expensive. Obvious metric, but idk, it feels a bit wrong. For coding agents especially, the expensive part is not tokens. It’s failed attempts, review time, “almost done” state etc. If Model A is 5x cheaper but needs 6 attempts, and Model B solves it once with fewer mistakes, the “expensive” model may be cheaper. I believe we should focus on performance benchmarks more. Not how cheap the tokens are, but how often the model get shit done.

I dropped this post into my J-space and let it simmer in the background.