Angad Singi

I sold Product Monk for $45K. Revealing my complete journey in tomorrow's edition. Find out here: allthingsnewsletter.com

@cooldeep_me None. Buy anything else. I own Interceptor 650 and battery issue is a very real painful problem. I want to get rid of my bike and probably buy anything else. 🤯

Should I go with Classic 650 or Continental GT?

Every four minutes, a worker dies in India due to unsafe working conditions. While enforcement of national occupational safety norms is one solution, <40% of Indian companies comply fully with these rules. Here's another solution: robots. Bengaluru-based Flo Mobility (@Flo_Mobility) just raised a $2.5M pre-Series A from Mela Ventures (@Mela_Ventures) and Arali Ventures to build the future of safe, compliant robotic construction automation. Their robots have: - Reduced workplace accidents by 30% - Moved 250,000 tonnes of material - 500 kg payload capacity per robot - Collectively traveled 50,000+ km - Achieved 99.9% uptime We visited their HSR Layout facility to see how they're getting it done. 00:00 - Flo Mobility in a Nutshell 00:25 - Entering Flo Mobility’s Workshop 01:18 - Flo Mobility’s Early Prototype 02:42 - Flo Mobility’s Latest Robot 05:24 - The Importance of Data in Robotic Automation 06:14 - Testing Flo Mobility’s Robot Out 09:05 - Flo Mobility’s Current Field-Deployment Model 10:18 - The Flo Mobility Story 19:26 - Exploring Flo Mobility’s Robotics Stack 20:37 - What’s Next for Flo Mobility?

@Kanishkhaai Would you be open to sharing with/teaching me the flow of chatgpt images -> Seedance Video Outputs? I am okay with a paid session as well..

GPT Image 2.0 on Chatgpt App PROMPT :- cinematic 6-frame grid (3x2), ultra-premium campaign for Nike Air Max 95 “Big Bubble” in deep berry pink + glossy rose tones STRICT consistency across all frames: same lighting setup, same color grading, same product color accuracy, same soft beige-pink gradient background, same luxury editorial tone lighting: soft diffused studio lighting + glossy specular highlights + subtle pink glow accents color palette: deep berry red, glossy pink, blush beige, warm ivory, soft gold reflections texture emphasis: matte mesh + glossy leather + translucent air bubble CAMPAIGN CONCEPT: “Soft Power in Motion” MOOD: feminine strength, confident, bold, effortless luxury, streetwear meets beauty campaign FRAME 1 — HOOK (surreal scroll stopper) Nike Air Max 95 floating mid-air, surrounded by flowing liquid pink silk + glossy fluid waves, subtle motion distortion, dreamy yet powerful composition FRAME 2 — TEXTURE (desire macro) extreme macro close-up of air bubble + layered material textures, translucent sole glowing with warm golden light, hyper-detailed reflections FRAME 3 — EMOTION (identity) confident female model wearing Nike Air Max 95, walking forward with attitude, soft wind in hair, minimal neutral outfit, strong posture, empowered presence FRAME 4 — MOVEMENT (performance) dynamic mid-air jump shot, motion blur trails, glowing air sole emphasizing bounce, high-speed capture feel, energy lines FRAME 5 — HERO PRODUCT (clean dominance) perfectly centered product shot, symmetrical composition, ultra-clean background, premium minimal aesthetic, sharp detail FRAME 6 — BRAND PAYOFF (iconic finish) cinematic silhouette of female figure wearing the sneakers, strong pink backlight glow, dramatic shadows, bold Nike energy, powerful stance STYLE: high fashion editorial + luxury product photography + cinematic sports campaign ultra-realistic, 4k, sharp focus, depth of field, premium finish, no distortion, no inconsistencies OPTIONAL TEXT OVERLAY (minimal, clean typography): “Soft Power” / “In Motion” / “Just Do It.”

Been building a system that turns any product URL into TikTok-ready videos. Gave it this necklace from Dillano Elan. The video above is what came out. Fully AI generated shots, voice, music, edit. I only gave it the link. Anyone here runs an ecommerce brand or does creator content? Would you actually post something like this on your TikTok? Curious what's missing. Drop a comment or DM if you've got thoughts. 👀

ugc & ad agencies, we are coming for you dm me if you'd like early access

@gangadhar__s Got it. Makes sense!

I don’t own a brand, but I do AI ads for a living, and my personal take is that no matter how good AI gets at video generation, one‑click ads will never be great to use. They’ll always lack something without any human involvement. Here video looks good but the copy sounds very basic.

We gave the system a product URL. This is what came out. 👆 No creative brief. No human direction. Just a URL. If you were a Shopify brand owner — would you post this on your TikTok, Instagram, YT? Trying to understand if this is good enough or if we're still too far off. 👀

Spent two weeks getting the video quality technically perfect. Every shot calibrated. Voice matched to brand personality. Captions styled. Music mood set. Then we watched it on TikTok. Nobody watched past 3 seconds. The feedback from the AI reviewer: "visually stunning, but studio quality is a skip signal in 2026. Users see high production and assume it's an ad." That hit. We optimized for quality. The problem was it was too polished. TikTok users scroll past anything that looks expensive. The lo-fi stuff gets watched. The stuff that looks like someone's phone camera gets watched. The stuff that looks like a brand hired a production crew gets scrolled. So now we're solving a different problem: how do you make AI-generated content feel raw enough to get views, without actually making it bad? The videos need an imperfect entry point. Something at the start that signals "real person, not an ad." The weird part: the technical quality is already there. The problem is making it look like it isn't. 🤯 Building the fix now. No idea if it'll work.

This video was generated from a product URL. Nothing else. No voice yet. Still early. Visuals are AI generated end to end. This is what the pipeline produces today — unfinished, but the direction is clear. 👀

30 videos. Me watching each one. Thumbs up or down. Claude finds the pattern and writes it into the quality prompt. I've never had to describe what I think looks good in this much detail before. Harder than building the pipeline. 😅

So now I have to put my own taste into words. What does "feels creator-made" actually mean? Handheld energy. Product in context, not floating. Imperfect framing. A line that sounds like something a real person would say, not a pitch.

Building an AI system that generates short videos. The quality system catches the technical stuff: black frames, audio cut off early, product missing from shot, distorted hands.

Built a video pipeline with 26 AI agents. More specialization, better output — made sense in theory. Videos were mediocre. Agents hallucinated without real product data. Context got lost at every handoff. QC approved things it shouldn't because it had lost track of what "good" meant three agents ago. Rebuilt. V2. Fewer agents, cleaner pipeline. Then realized V2's entire structure was built around the model we were using. Two-clip limit. 12-second max. Prompts joined with ". Then:" because there was no multi-shot support. Switched models. V2 wasn't just outdated. The foundation was wrong. On V3 now. From scratch. Spent a week just understanding what the new model could actually do before writing any code. Found things that weren't in the docs. Building around what's real this time. Two failed versions make the starting point a lot clearer. 😅

Kling 3.0 has a 512-char limit per shot prompt. Not in the API docs. I was sending 840-char prompts. Got "Unprocessable Entity" with zero details. Took the whole day to find it. 😅

Been using github.com/garrytan/gstack by @garrytan for a week now. It's a set of skills for Claude Code. QA testing, code review, design audits, shipping workflows, browser automation. All inside the Claude desktop app I already live in. The stuff I use most: /qa finds bugs and fixes them in a loop /review does a staff-engineer-level code review before I push /design-review catches visual issues I'd never notice /plan-ceo-review rethinks the product direction before I commit to building /plan-eng-review locks down the architecture before I write code /browse runs a headless browser without leaving Claude. Life saver. It's one of those tools where you don't realize how much you relied on it until you try working without it. Thank you @garrytan for building this 🙏

This response from Claude changed how I think about building with agents. 🤯

Spent a day trying to fix the gap between design and code. Tested three tools on the same screen at the same time. Paper.design: everything is real HTML/CSS, not a proprietary format. Built by the guy who made Radix UI. His MCP lets Claude read and write directly to the canvas. You change a color in the design, the code changes. No translation layer. Google Stitch: free, upload a screenshot, get a full redesign in 30 seconds. Gave it 11 screens. It redesigned every one. Looked better than mine 😅 A Claude plugin I didn't even know I had: already installed, quietly improving everything Claude generated. No setup. Just better output. None of them solved the whole problem. The output looked good. But I couldn't explain why it still wasn't quite right. And getting it there took 10 rounds of back and forth anyway.