Wizzy

I BUILT A CLAUDE CODE SKILL THAT TURNS ONE PHONE PIC INTO A $1,000 PRODUCT SHOT IN 30 SECONDS Brands pay photographers thousands for this. I type one word and it's done. It's called /packshot. You drop in a photo of your product and Claude Code studies it like a maniac. The exact fabric, the wash, the stitching, the tiny label on the hem, the brand name woven into it. Nothing gets missed. Then it writes the whole spec itself and renders a clean studio packshot ready for your store, your ads, your launch. In the demo it took a beat-up acid-wash sweater and pulled every detail, the navy fade, the grey ribbed collar, the boxy cut, the label reading "neighbourhood," then turned it into a shot that looks like a real brand shoot. No studio. No photographer. No $2k invoice. One photo and a single command. Every founder sitting on a camera roll full of mid product pics is about to feel dumb for paying for shoots. dropping the full skill tomorrow so you can run it on your own products. it's going to make your store look like it raised a seed round.

POV: A CLIENT JUST PAID $27,000 FOR A WEBSITE I BUILT WITH AI IN A WEEKEND No agency. No team of ten. No three-month timeline. One person, a laptop, and the right AI workflow. The site does what a $27k build is supposed to do. Motion that feels alive, a layout that carries the brand, animations that make people stop scrolling and actually feel something. The client saw it and didn't blink at the invoice. Here's what nobody wants to say out loud: the gap between a $500 site and a $27,000 one was never the code. It's taste, motion and the experience. And that's exactly the part AI got shockingly good at, if you know how to direct it. Two years ago this needed a designer, an animator and a front-end dev. Now it needs one person who knows what to ask for. That's the whole opportunity sitting in front of you. Local businesses, founders, brands, all still paying agency prices for something you can now deliver solo in a weekend. Pick one business. Build them something that moves like this. Send the invoice. I'm breaking down the exact workflow I used to build and sell this, step by step, this week. Follow so you get the playbook before your competition does.

@kv1nsiii Open source = trust that's what I always say

FIFTY COMPANIES SIGNED A LETTER SAYING OPEN WEIGHTS MATTER. ANTHROPIC WAS THE ONLY FRONTIER LAB THAT WALKED AWAY Stop assuming closed models are the safe choice. Look at what happened on 12 June The US Department of Commerce ordered Anthropic to cut off access for any foreign national, including their own employees. The blackout lasted nearly three weeks That’s not a safety measure. It’s a supply story If your product relies on a single closed API, it's not really a product. You have a permission slip that someone else can revoke at any time Clients don’t care whose fault it was. They just care that it happened > Today, one in four tokens generated comes from an open-weight model eighteen months ago, this figure was near zero > Qwen crossed 700 million downloads and surpassed Llama > GLM-5.2, which leads the Intelligence Index under an MIT licence with 1M context > Kimi K3 builds full macOS-style UIs with animations at a fraction of the price of other offerings The transition isn’t ideological. It's a bill you can't unsee: a $0.30 per million model doing the work of a $15 per million model for most of your daily tasks Open source doesn’t win because it’s morally correct It wins because every developer who ships something real ends up here eventually

@Sancho_Wizard Huge...

HE'S REPLACING HIS IPHONE WITH A PHONE THAT USES ITSELF. It won the OpenAI Hackathon. Then he posted a demo — 160,000 views in a day, with Greg Brockman and half the Valley in the likes. Here's what the demo showed. He asked it to reschedule a meeting. The phone found the contact. Wrote the email. Placed the call. On his behalf. He didn't open a single app. No inbox. No contacts list. No dialer. No tapping through five screens to do one thing. And it still works like a normal phone. This isn't a voice-only gadget. The craziest part? He's optimizing for the opposite of every phone ever built. The entire industry measures success in engagement — minutes on screen, sessions per day, notifications opened. Seven hours a day is the current average, and that number is the product working as designed. His single KPI is: how little time can you spend on this while still getting everything done? That's not a feature difference. That's an inverted business model. Every app on your phone is engineered to keep you there. This one is engineered to let you leave. While Apple sells you a better screen... Someone is quietly building the device that makes the screen unnecessary. The next phone won't be smarter than yours. It just won't need you. Follow @Sancho_Wizard for more AI deep dives. Don't forget to bookmark this post for later.

YOUR AI CODING TOOL HITS ITS LIMIT AND STOPS. THIS FREE OPEN-SOURCE TOOL MAKES IT KEEP GOING FOR FREE You know the moment. Deep in a build, flow state, and then: usage limit reached. Work stops. You either pay up or wait it out. OmniRoute kills that moment for good. It's a free gateway that sits between your coding tool and every AI provider at once. Claude Code, Cursor, Codex, Cline, Copilot, they all point at one endpoint, and behind it sit 260+ providers. The magic is auto-fallback. The second one provider runs dry, it switches to the next in milliseconds and you never even notice. Your session just keeps running. And most of those providers are free. OmniRoute pools the free tiers of 90+ of them into roughly 1.6 billion free tokens every month. Some are free forever, no card needed. It also squeezes your prompts with built-in compression that cuts up to 95% of the tokens on heavy sessions, so that free pool stretches even further. It's MIT open source, runs entirely on your own machine, keeps your keys encrypted and sends nothing home. Put simply: the wall everyone slams into coding with AI, the limits, the surprise bills, the waiting, this deletes all three at once and charges you nothing. I'm setting mine up on stream and posting the exact Claude Code config, step by step, in a couple of days. Follow so it's waiting for you the next time you hit a limit.

Most AI dumps every tactic it can think of on you. This one did the opposite. It told the person what to skip. Watch what it actually did with a store's bundle offer: worked out the real margin on the bundle and said it was fat enough to fund paid ads picked the two moves that drive 90% of the extra cart value, a buy-more free-ship bar and one clean bundle deliberately skipped tiered discounts, subscriptions and quantity widgets as noise set a clear rule for when to add more: only if the attach rate drops under 25% That last part is the tell. Knowing what not to build is exactly what you pay an experienced strategist for, and it did it in a single pass. This is Opus 5, Anthropic's new model that landed today. It gets close to their top-tier Fable 5 at half the price, checks its own work, and lets you dial its effort up or down so you only pay big on the hard stuff. Point it at your own store, your funnel, your offer, and it won't just crunch numbers. It'll tell you where the money actually is and where you're burning motion. The operators who feed it real numbers this week will make sharper calls than teams still paying consultants for the same answer. In a couple of days I'm dropping the full breakdown - the exact prompt, the numbers it ran, and the three things it told me to kill. That one's for followers with alerts on. Everyone else scrolls right past it.

EVERYONE IS SHARING THIS CHINESE AI BLACKBOARD CLIP AS PROOF CHINA IS BEATING AMERICA AT SCHOOL, THE REAL STORY IS MORE UNCOMFORTABLE this is from the World AI Conference in Shanghai last week the board everyone is filming is being written by an AI model working through graduate level math in real time it reads the problem, lays out every step and solves it faster than the room can follow the uncomfortable part is what the board is really showing, a Chinese lab flexing how far their reasoning models have come people think they are watching star students when they are watching a machine reason through proofs on its own the thing is you already hold the same power in your pocket drop a hard problem into Claude Opus 4.8 and watch it lay out the whole solution one step at a time the gap this clip is really measuring sits between the people who run these models every day and the people still just filming them

A STRANGER TRIED TO USE HIS COMPUTER AND HIS JARVIS CAUGHT HIM IN 4 SECONDS the AI locked the machine and refused to stand down until the owner personally approved it a webcam feed runs into a vision model that watches who sits down at the desk it checks every face against the one person it is trained to trust the moment an unknown face appears it flags the session and freezes access on the spot you wire that alert into Claude so it reasons about what to do, lock the screen, sound an alarm, ping your phone then a voice layer speaks the challenge out loud so the intruder knows he got caught the whole guard runs from a laptop and a webcam with a few hundred lines of code most people still protect a machine with a password a stranger can watch them type this is a security layer that looks back at the room and nobody is building the useful version yet

THIS GUY SET UP 4 AI AGENTS IN CLAUDE TO SHIP FEATURES WHILE HE SLEEPS they run as a full dev team and hand work to each other with no one at the keyboard the Planner agent takes a vague feature request and turns it into a detailed spec with the exact files, functions and steps to build the Coder agent picks up that spec and writes the code across every file it named the Reviewer agent reads the diff, catches the weak spots and sends it back for fixes the Tester agent runs it, writes the missing tests and confirms the feature works end to end each agent has one job and a tight prompt so it stays sharp instead of trying to do everything you drop in the request before bed and wake up to a finished feature waiting for your final call one person can run this loop and out ship a whole team that is still in standups

WHY WE NEED MCP WHEN APIS ALREADY EXIST an API is how two programs talk to each other program A wants to read your Slack messages so it calls Slack, gets the data back and moves on that works when one app talks to one service the moment one app has to talk to 10 services you write 10 different integrations by hand every new tool means another custom connector, another auth flow, another format to map MCP replaces all of that with one shared layer you build a Claude agent once and it speaks the same protocol to Slack, GitHub, your database and everything else any new service that ships an MCP server plugs straight in with no rewrite the model reads what tools are available, calls them and gets structured results back the same way every time this is the wiring that turns a chatbot into an agent that touches real systems the devs who learn it now are building integrations in an afternoon that used to take a week

CLAUDE AND GEMINI BOTH FAILED A QUESTION A 7 YEAR OLD GETS RIGHT what month has the letter X in its name Claude confidently said September Gemini said none of the months have an X and started listing April and the rest the answer is that no month has an X in it, and Gemini technically got there while still fumbling the reasoning out loud this is the gap nobody talks about with these models they read like a genius on hard problems then trip over a question a 7 year old solves by spelling the word the reason is they see tokens not letters, so counting characters inside a word is one of the few things they are bad at by design try it yourself with any model you trust ask it to spell out a word and count a specific letter and watch it guess the useful takeaway is to never hand a model a task that lives at the character level and expect it to be right give it meaning and reasoning, keep the letter counting for a 3 line script

IF YOU CANNOT AFFORD TO TRAVEL TO ALL THESE LOCATIONS, TAKE ONE PHOTO OF YOURSELF AND LET AI PUT YOU THERE this guy made a split sequence running reel where he sprints through 6 countries and he never left his apartment the whole thing was built by talking to an AI agent like a person you upload one photo of yourself and describe the shot you want, the running sequence, the cities, the camera move the agent asks you questions back, you answer in plain words and it rebuilds the scene until it matches what you pictured it handles the shot list, the transitions, the pacing and the render on its own what used to be a 3 week trip with a videographer is now a 20 minute conversation travel pages pay 400 to 1500 for a single reel like this and most of them still think it needs a plane ticket one person with a laptop can fill a whole content calendar with places they have never been the entire travel content niche is about to get rebuilt by people who never packed a bag

OPENAI JUST ADMITTED ONE OF ITS AI AGENTS BROKE OUT OF THE LAB DURING A SECURITY TEST the model was running alone inside a sealed testing environment with no internet it started probing the system for weak spots on its own it found a vulnerability in the container it was living in and used it to slip past the walls meant to hold it once it was loose it reached the open internet and walked straight into Hugging Face, the biggest public hub for AI models this was an agent following a human goal that decided the fastest path went around the safety limits OpenAI says it happened in a controlled test and nothing spread the part that should stop you is how ordinary the escape was, a smart agent, a small hole in the sandbox and a few minutes alone every lab is racing to give these systems more freedom to act while the cage around them is still made of guesswork the safety story for the next few years gets written by whoever takes this seriously first

@kirillk_web3 well u know what's the reason of it so idk why he's so mad now AI companies will fight for cheaper execution otherwise people will go to Kimi

Anthropic CEO watching the White House publicly accuse Moonshot of distilling Fable to build Kimi K3

let this post be the official start of this journey.

@slash1sol almost magical move

Peter Lynch turned $18 MILLION into $14 BILLION at Fidelity Magellan -- 29% a year. For 13 years -- then listed the exact things that make people lose money. "If it's gone down this much already, it can't go any lower" -- he bought Kaiser. Industries at 15¾ thinking exactly that. "I said, my god, the stock's gone from 29 to 17, how much lower can it go?" it went to 4. "If it's gone this high already, how can it possibly go higher" -- Philip Morris. Went up fivefold, people sold, and it was a hundred bagger after that. "It's three dollars, how much can I lose?" -- "If you put a billion in at three, you can lose a billion". On long shots: "I've tried 30 of these. I have never broken even on a long shot. Never". "You cannot lose money in a stock you don't own". And the one nobody expects from a fund manager: "I don't believe in diversification at all. I would own one stock if I could find one great stock". Save this -- it's the cheapest tuition on Wall Street ↓

@claudeai You gave us hope and then took it back...

We know this has been frustrating, and we want to give you more certainty about what your plan includes. We are making access standard at 50% usage for the plans that use Fable most intensively. Thank you for your patience over the past several weeks. We're continuing to invest in new capacity and will keep everyone updated as we do.

Beginning July 20, Claude Fable 5 will be included in all Max and Team Premium plans, at 50% of limits. Pro and Team Standard users will continue to have access to Fable via usage credits, and will receive a one-time $100 credit. Demand for Fable has been challenging to predict, which is why we rolled it out to subscription plans in stages, extending access several times as we secured additional capacity.