Ash Minhas

@mikefutia YOUTUBE

This YouTube AI agent is absolutely wild 🤯 It scrapes top YouTube videos in your niche, analyzes them with AI, and extracts every creative insight you need. All inside n8n + Airtable. Perfect for DTC brands & agencies who need to know what's working on YouTube before they create ads or content. Here's the problem: Manual YouTube research takes forever. You're watching dozens of videos, taking notes on hooks, trying to remember what angles worked. And by the time you analyze everything, you've wasted hours. This n8n automation solves it: → Enter a keyword (e.g., "rucking", "cold plunge", "protein powder") → AI scrapes top YouTube videos with transcripts automatically → Writes all videos to Airtable with views, likes, published date → Gemini analyzes each transcript and extracts: Hook, Angle, Theme No manual watching. No note-taking. No wasted hours. What you get in Airtable: → Video URL, title, channel, performance metrics → AI-extracted hooks (what grabbed attention) → Angles (the unique claim or perspective) → Themes (the core narrative structure) → Full transcripts for deeper analysis Built 100% in n8n. Want the complete n8n template + Airtable base? > Comment "YOUTUBE" > Like this post And I'll send it over (must be following so I can DM)

@mikefutia BANANA

Nano Banana + n8n is legitimately insane 🤯 Google's Nano Banana creates studio-quality static ads. But manually generating variations one-by-one takes forever. This n8n automation generates 1,000+ ad variations in minutes. Fully automated. Perfect for DTC brands & media buyers who need fresh creative for testing without hiring designers. Here's the problem: You need 50+ ad variations to test angles, but Nano Banana only outputs one image at a time. Manually uploading and tweaking prompts for each variation kills hours. This n8n automation solves it: → Upload ONE product image via n8n form → OpenAI Vision analyzes your product automatically → AI generates custom image prompts (you choose quantity: 50, 100, 1000+) → Nano Banana creates static ad images in bulk → All images auto-stored in Box for instant access No manual prompting. No designer bottlenecks. No waiting on agencies. What you get: → Hundreds of unique ad variations from one upload → Different angles, backgrounds, compositions → Production-ready static ads → Perfect for testing creative on Meta/TikTok Built 100% in n8n. Want the complete n8n template? > Comment "BANANA" > Like this post And I'll send it over (must be following so I can DM)

Check out my blog for a detailed description of my experience thinkscientist.com/vibe-coding-wi…

If you live in #NewJersey and use the @PATHTrain then join the public beta here -> testflight.apple.com/join/BCGGUqpU

I used GPT-5 and Cursor to add some features to an iPhone App.

Can't wait to get access to this...

Had a play with Google's experiment Opal and made a kid's bedtime story generator. Looks eerily similar to tldraw in the editor. Video sped up (2x) opal.withgoogle.com/?flow=drive:/1…

2/ Improve consistency in multi-turn conversations. Caching isn’t just a developer trick—it’s essential for scale! Have you built with prompt caching? How much latency did you cut? #LLM #LangChain #AI #PromptEngineering ibm.com/think/tutorial…

1/ ⚡️ Prompt caching is the unsung hero of LLM-powered apps! Instead of paying and waiting for new responses every time, cache past completions and instantly serve repeat results. This IBM tutorial walks you through SQLiteCache: Save on costs & Speed up your LLM App

3/ Read all about the models on the IBM Github Page github.com/ibm-granite/gr… #TimeSeries #AI #Forecasting #IBMGranite

2/ It's great for for both few and zero-shot forecasting and it can easily be fine-tuned for multivariate datasets too.

1/ 📈 Want next-level retail demand prediction? This tutorial takes you through using a model from IBM's Time Series Foundation Model Family (TinyTimeMixer) It's open-source, really tiny (<1M params) and performs well—no billion-parameter beast needed ibm.com/think/tutorial…

This is wild! @tldraw has blown my mind with how far down the rabbit hole their infinite canvas has got. @steveruizok gives a great demo here youtu.be/1C2TdPkj6aQ?si…

@Lane8music what was the George Fitzgerald burns remix you played last night at the Brooklyn mirage. I can’t find it! It was incredible!

@eurydice13 @Ocado i hope theres some chocolate in the basket too :)

I wonder what a cynic might say to this.... 🤔 @Ocado

@johncutlefish @swardley Would love to see what’s inside

@allenholub No need to apologize, your expansion and articulation was spot on. Thanks for everything you share. It’s inspiring :)

@thinkscientist Sounds perfect. Sorry I misunderstood.

What’s the valuable outcome on each of the stories in your backlog?

@allenholub No it was a retweet of your original with a provoking question for others. I agree with you. Teams that I lead go out and talk to users. Meet them where they are, look at what they are trying to do, then prototype and validate their understanding

@thinkscientist Maybe your comment is coming from a misunderstanding of where stories come from. We capture them from conversations we have with our users. *We* do not make them up. They reflect user needs. 2/2

@allenholub And how do you find out what the valuable outcome is without measuring and testing? Rather than stories or requirements, why not hypothesis?

@thinkscientist Don't understand the question. Your user or customer is using your system to achieve some objective or result that they consider valuable. The "valuable outcome" is whatever the user considers valuable.