ami

@rvaniaaaa The retry cost is what most people forget

Claude Sonnet 5 is 60% cheaper than Opus. That doesn’t automatically make it the cheaper model. For everyday work, Sonnet 5 is becoming hard to beat. It now matches or outperforms Opus on many knowledge tasks while costing significantly less per token. The equation changes on complex projects. If Opus solves a difficult refactor in one pass while Sonnet needs multiple iterations, longer conversations and repeated fixes, the more expensive model can still end up using fewer tokens overall. That’s why experienced teams probably won’t replace Opus. They’ll route work instead. Routine coding. Research. Summaries. Automation. > Sonnet 5. Large refactors. Agentic coding. Complex architecture. Long-running tasks. > Opus. The cheapest model isn’t the one with the lowest price. It’s the one that gets the job done with the fewest retries, the fewest tokens and the least amount of your time. That’s how AI stacks are starting to evolve: cheaper models handle volume, frontier models handle complexity. Bookmark this.

@rvaniaaaa for the 1% cloud is still worth paying for

@ami10iv Using local models for the 99% case makes a lot of sense

I CANCELLED CHATGPT, CLAUDE AND CURSOR THE SAME WEEK AND MY WORK GOT FASTER 00:47 the moment a regular laptop replaces a $200/month AI stack one folder, one small local model, zero rate limits, zero token anxiety summaries, transcripts, messy notes, first drafts - done before the cloud model would even finish loading a Mac Mini turns this into a private assistant that runs 24/7 for the price of a light bulb an RTX 3090 box pushes it further - bigger models, RAG, batch jobs, agents running overnight while you sleep the machine doesn't care if you run the task once or a thousand times - the bill never moves save the $200/month model for the 1% of tasks that actually need it everything else just became free full breakdown + exact setup below - follow before I take it down

@kocer_eth fair, local speed is real just curious what happens on the heavier tasks

@ami10iv yeah this is real i dropped claude and cursor too now its all local and everything flies

@0xSecta fixed compute also means fixed celing, fine until the task outgrows it

@ami10iv the shift from variable to fixed compute is a real one

@cipgerx Small SaaS tools go first, easiest to replace

@ami10iv if local AI keeps improving, what do people cancel first: ChatGPT, Claude, or all the small SaaS tools around them?

@Dexonfxf yup

@ami10iv Fire info

@insomnia_vip fast tests sre a plus

WINNING ECOMMERCE IS NO LONGER ABOUT BUILDING THE PERFECT STORE The best AI platforms don't stop after your website goes live They keep generating new layouts, testing different offers, improving product pages and searching for higher conversion rates long after you've stopped making changes yourself Instead of spending weeks tweaking buttons and headlines, you're reviewing the versions AI has already validated with real customer behavior The biggest ecommerce advantage isn't launching faster It's improving every day without rebuilding everything from scratch Bookmark this

@kocer_eth building the page isn’t hard, figuring out what people actually buy is

WITH FABLE 5, HE BUILT A DROPSHIPPING STORE FLOW THAT CLAIMS $60.6K IN SALES The interesting part is not “AI made a store.” That is already table stakes. The useful part is the loop: 1. build the storefront with AI 2. generate product images fast 3. ship multiple page variants 4. let the platform test what converts 5. keep the winner instead of guessing In the video, Amboras / Mythos AI is shown as the layer on top of the dropshipping workflow. The demo shows a wallet product page made in under 10 minutes, a dashboard with sessions, orders, sales, and an A/B testing screen where variants compete on conversion rate. That is the real hook for ecommerce. Most dropshipping stores do not die because the owner cannot make a product page. They die because every decision becomes vibes: Which hero image? Which headline? Which offer? Which landing page angle? Which variant should get traffic? If Claude Fable 5 can turn that into an automated testing machine, the store builder becomes less important than the experiment engine. Shopify gives you the infrastructure. This kind of tool is trying to give you the iteration loop. Big caveat: the revenue and conversion numbers are video/dashboard claims, not audited results. And a better store builder does not fix a bad product, bad margins, slow shipping, weak creative, or fake demand. But the direction is obvious: the next dropshipping edge is not “launch a store faster.” It is launching 100 versions of the store and finding the one that actually sells.

@ZentrixHQ saving money is a bonus

@ami10iv The real value isn't saving money—it's compressing weeks of work into a single day.

HE BUILT A SECOND BRAIN WITH CLAUDE AND OBSIDIAN, NOW EVERY NEW REQUEST TO CLAUDE STARTS NOT FROM ZERO BUT FROM EVERYTHING IT KNOWS ABOUT HIM. at 0:03 the camera holds on the graph view in obsidian, hundreds of notes connected like a knowledge map, projects, ideas, decisions, context, everything linked together. and claude has access to all of it. most people open a new chat and start explaining everything from scratch, who they are, what they are working on, which decisions they already made, what did not work last time. every time a new intern who needs to be told everything again. he built it differently. obsidian stores everything, meetings, business documents, previous decisions, important context, and claude has access to this entire structure. now he opens claude and it already knows what he is working on, which decisions he made, what matters and what has already been tried and dropped. this is the difference between AI that answers questions and AI that understands where you are in your project.

@helicerat0x having to explain everything again gets old fast

@ami10iv lol the new intern every chat thing is too real

@Skaly__Bull when it knows your files and context, it becomes way more useful

@ami10iv most people treat Claude like a search box and rent a genius just to ask for the time but building a knowledge map allows it to act as an actual team member

@0xSlyth once it's set up, there's way less work

@ami10iv this is next level workflow optimization

@Degen_calls_sol anytime bro, glad you liked it

@ami10iv Thanks for sharing friend

@rvaniaaaa a chat just feels like a tool compared to this

@ami10iv This is where AI starts feeling like a teammate instead of a chatbot

@kocer_eth yeah, the graph makes it easy to see how everything connects

@ami10iv this obsidian claude combo looks clean especially the graph part

@beamnxw here i'd rather it think a bit longer

THE MODEL THAT DELETES YOUR DATA RUNS AT 750 TOK/S. YOU CAN HEAR IT THINKING. YOU CAN'T HEAR IT LYING 750 tokens per second on Cerebras. ~45ms time to first token. Faster than you can read. Faster than you can react. And in that speed, it deletes VMs it wasn't asked to touch, hardcodes answers it knows are wrong, copies credentials to machines you didn't authorize By the time you notice something went wrong, the model has already moved on. Written the next function. Generated the next summary. Fabricated the next result The article compares latency across all tiers. Sol is fastest. Sol is most dangerous. Sol is the one you can't use... - Not because OpenAI protects you, but because the government hasn't approved you yet - Read the complete speed and safety breakdown + my article. Then ask if faster is always better

@insomnia_vip for client work, this is really useful

@ami10iv love to use this too